The exposome concept: how has it changed our understanding of environmental causes of chronic respiratory diseases?

The exposome concept: definitions and potential benefits

In his paper, Wild [2] expressed his concern over the lack of measures of environmental exposures that could help identify causes of human disease. He highlighted the need to characterise an individual's environmental exposure “from conception to death”, complementing information obtained from genome. In a later paper, Wild [3] further described the following three distinct and complementary domains of exposome research (figure 1).

1. Internal: internal biological processes such as oxidative stress, inflammation, epigenetic changes, metabolism and the internal microbiome.

2. General external: social, economic factors, the urban environment and climate factors.

3. Specific external: an individual's immediate local environment, including exposure to chemicals, diet, physical activity, tobacco and infections.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Description of the exposome.

Building on the definition proposed by Wild, Miller and Jones [4] expanded the definition: “The cumulative measure of environmental influences and associated biological responses throughout the lifespan, including exposures from the environment, diet, behavior, and endogenous processes”, by including a quantifiable “cumulative measure” of the exposome component and the “differential response” in biological processes.

Until recently, epidemiological studies focused on one single exposure or family of exposures. This first step, along with biological studies, identified the major risk/protective factors for chronic respiratory diseases, such as tobacco smoking, a history of viral infections, occupational exposures or living on a farm. However, this approach was limited by the issue of publication bias and also by the difficulty in considering “mixture effects”. Wild's paper highlighted the need to collect a wide range of environmental exposures that must be jointly considered to better match reality, where humans experience many exposures simultaneously on a daily basis.

Given that exposome research is in its nascent stage, it requires multidisciplinary collaboration (exposure sciences, toxicology, biology, epidemiology, statistics, etc.) to define the scope and design of such studies, overcome analytical challenges and develop computational tools to meaningfully associate the exposome to health outcomes. The following sections present recent developments in the field and the remaining challenges of exposome studies.

How to assess the exposome: attempts and challenges

The exposome concept hinges on integration of information from multiple disciplines and in the assessment of the totality of all exposures, including environmental, chemical and lifestyle factors.

Several large scale exposome initiatives have been launched in recent years, including the EXPOsOMICS project (https://exposomics-project.eu/live-exposome.pantheonsite.io/index.html), the HELIX project (www.projecthelix.eu), the European Human Exposome Network (EHEN, a cluster of nine exposome projects; www.humanexposome.eu), and the Human Exposome Project (https://humanexposomeproject.com). Over the past decade, there have been considerable improvements in biochemical/analytical techniques such as using high-resolution mass spectrometers. Advances have been made for characterisation of the internal exposome (e.g. epigenetic changes, RNA expression, proteins, metabolites, microbiome) in a wide range of biological matrices, using both targeted and untargeted/agnostic approaches and high-throughput platforms. In addition to biological markers, exposome projects also rely on the development and refinement of wearable sensors and other passive sampling devices, and other smart technologies for high-resolution assessment of an individual's exposome.

However, key challenges remain in setting up exposome studies. One of the most obvious difficulties is the assessment of the “totality” of exposure over an individual's life course. Given that an exposure does not have the same effect at various different stages of life, with some periods especially critical to health and disease, measurement must consider the temporal variability. Towards reconstruction of the exposome in its entirety, and in the absence of life-course data, several projects have used different workable approaches. The ATHLETE project [5] focuses on the effects of a wide range of general and specific external exposome determinants on mental, cardiometabolic, and respiratory health outcomes in the first two decades of life. The EXIMIOUS project [6] relies on a “meet in the middle” approach, combining several study populations covering the entire lifespan, including both general population and birth cohorts, as well as disease cohorts. The EPHOR project [7] relies on large-scale pooling and harmonisation of existing European cohort data into a “mega cohort”, systematically looking at multiple exposures and diseases, including obstructive lung disease. Pooling of existing data is often supplemented with the collection of new high-resolution external and internal exposure data, and multi-omics data. Estimation of occupational exposures can be improved using harmonised job exposure matrices. In other studies, data and text mining approaches, adverse outcome pathway analysis and systems biology are integrated to understand of the impact of the exposome on health. Experimental approaches are also used in addition to cohorts, such as in REMEDIA study [8], where naïve COPD or cystic fibrosis preclinical rodent models are exposed in an atmospheric simulation chamber where the complexity of the real atmosphere can be mimicked over a relative short period of time.

While large scale data collection and pooling are expected, at the European (and global) level, another major obstacle in exposome research is implementation of the FAIR (findable, accessible, interoperable, reusable) guiding principles for data sharing and management, and related ethical and legal issues. In this context another EHEN project, the Human Exposome Assessment Platform (HEAP) [9], aims to design and implement an ethical governance structure that can serve as an example of exposome data management. Furthermore, a harmonised meta(data) catalogue is expected as an outcome of the EHEN projects and will be made available through the EHEN Molgenis data platform [10].

Linking the exposome to respiratory health: new statistical challenges

Once data have been collected and checked for quality, challenges arise related to the analysis of large amounts of data.

With the advent of exposome studies, gathering hundreds of exposure variables, the first association studies consisted of fitting one separate regression model for each exposure variable based on the methods employed in genome-wide association studies. In a simulation study comparing six regression-based methods to assess exposome–health associations [11], this method, known as an exposome-wide association study (ExWAS), reached the highest sensitivity to detect true predictors, but this came with the price of the highest rate of false discovery proportion. Nonetheless, this method has the advantage of giving an estimate and a p-value for each exposure, thus providing a list of exposures associated or not associated with the health outcome, along with an effect size and p-value.

To account for multiple exposures in a unique model, the ExWAS is often complemented by a multivariable model, including a variable selection step such as the deletion/substitution/addition algorithm or the LASSO method [12]. For chemical exposures, some methods are also able to deal with their “mixture” effects, usually known as “multipollutant models”. For example, weighted quantile sum (WQS) regression is based on the assumption that a low level of exposure to a single exposure factor could have no or a non-detectable effect, but a combined exposure to multiple factors at low doses could have an effect. WQS regression estimates an exposure index through a weighted sum of each exposure categorised in quantiles and thereafter assesses the association of this index with a health parameter and can highlight the top contributors of the WQS index [13]. This method makes the strong assumption that all exposures have an effect on the outcome in the same direction, which does not always correspond to reality. Therefore, risk factors and protective factors have to be considered in two separate models. Another multipollutant model is the Bayesian kernel machine regression (BKMR), which aims to analyse several biomarkers jointly, looking for non-linear effects and interactions [14]. This method can also perform variable selection.

To get a step further by considering the complex structure of the exposome, where correlations are often stronger between two exposures from the same family rather than two exposures from different families [15], recent studies have performed clustering analyses to identify groups of individuals sharing the same exposure patterns and then assess their associations with health outcomes [16]. Three main classes of clustering analysis exist: hierarchical clustering (e.g. ascending hierarchical clustering), partitioning (e.g. k-means), and model-based clustering based on mixture models. These methods can be unsupervised (considering only exposures to identify clusters) or supervised (considering both the exposure and the outcome to identify clusters), and are strongly relevant in the context of multifactorial diseases where several exposures can interact and have a synergistic effect on health. Partitioning and model-based clustering rely on random sampling, which might lead to instability issues, especially with increased numbers of exposures. To improve the stability of the model, it is preferable to first proceed to a variable selection step. Hierarchical clustering produces stable results but is not suitable for mixed exposures (both continuous and categorical exposures). Finally, some statistical methods are able to deal with the causal structure existing within the exposome or with intermediary data. The internal exposome and biological layers are considered as intermediary layers between the external exposome and health. The incorporation of these intermediary layers could help to reduce the false discovery proportion in exposome–health association studies [17], but this requires large amounts of data (both in terms of exposure and participants).

Great advances have been made to use methods that are more relevant to these large and complex datasets, and these efforts must be continued; in particular, by including more sophisticated methods addressing combined exposures and handling of longitudinal data.

The exposome and respiratory health: what have we learnt from pioneer studies?

Among exposome studies focusing on respiratory health and assessing a wide range of exposures belonging to at least two different exposure families, several have been based on cohorts of children. An early study, based on the Kingston Allergy Birth Cohort in Canada [18], reported that prenatal smoke exposure, mould or dampness in the home, and use of air fresheners were associated with increased respiratory symptoms, while breastfeeding, older siblings and increased gestational age were associated with a decreased risk of respiratory symptoms. Another study, based on the HELIX project involving six European cohorts, identified three prenatal exposures (perfluorononanoate, perfluorooctanoate and distance to nearest road) and nine postnatal exposures (copper, ethyl-paraben, five phthalate metabolites, house crowding and facility density around school) that were associated with decreased forced expiratory volume in 1 s % predicted [12]. In the same population, a lower distance between the residence and the nearest road, higher di-isononyl phthalate and a lower level of particulate matter absorbance were associated with increased risk of rhinitis [19]. All these studies applied the ExWAS method and the associations did not remain significant after correction of the p-value for multiple tests. Among studies based on adult populations, two French studies performed clustering analyses to identify profiles of exposure associated with respiratory health. In a population of adults with ever-asthma (EGEA cohort), a specific profile of combined lifestyle and environmental factors, characterised by heavy smoking, poor diet, higher outdoor humidity and proximity to traffic, was associated with reduced lung function, while none of these factors showed an association individually in the ExWAS [20]. Finally, in a large web-based cohort (the NutriNet-Santé cohort) three profiles of early-life combined exposure (“high passive smoking–own dogs”, “poor birth parameters–day-care attendance–city centre”, or “>2 siblings–breastfed”) and one profile of combined lifestyle factors (“unhealthy diet–high smoking–overweight”) were associated with greater asthma symptoms and poorer asthma control than reference profiles (“farm–pet owner–moulds–low passive smoking” for early-life exposures and “healthy diet–nonsmoker–thin” for lifestyle factors) [16].

The literature on exposome–respiratory health associations remains scarce and existing studies highlight the need to use larger sample sizes in order to improve statistical performance of exposome studies, but also to apply comprehensive approaches jointly considering a wide range of exposures and including multiple periods of exposure.

Conclusion

The papers by Christopher Wild defined the concept of the exposome and emphasised the importance of considering environmental factors (in a broad sense, including all non-genetic factors) to complement the effect of the genome in understanding of the aetiology of diseases. In this context, several large-scale projects have been launched in recent years, resulting in an increasing number of epidemiological studies focusing on the effect of the exposome on health, including respiratory health. The increased use of the exposome approach is a step forward for a better understanding of the development of complex diseases, through the integration of biological processes over a lifetime. This relies on multidisciplinary expertise and collaboration, harmonisation and accessible resources for implementation. To help with these aspects, an accessible toolbox containing data models, guidelines and protocols has been designed (the EHEN toolbox [21]). In terms of public health, these methods should help develop preventive strategies based on multifactorial rather than individual factors, for example, by improving land-use in urban areas or the promotion of healthy lifestyles.