Enhanced exposure assessment and omic profiling for high priority environmental exposures in Europe

Informations projet

EXPOSOMICS

N° de convention de subvention: 308610

Projet clôturé

Date de début 1 Novembre 2012

Date de fin 30 Avril 2017

Financé au titre de

• Specific Programme "Cooperation": Environment (including Climate Change)

Coût total € 11 649 540,40

Contribution de l’UE € 8 748 628,80

Coordonné par

IMPERIAL COLLEGE OF SCIENCE TECHNOLOGY AND MEDICINE
United Kingdom

Final Report Summary - EXPOSOMICS (Enhanced exposure assessment and omic profiling for high priority environmental exposures in Europe.)

Executive Summary:
The exposome concept has been proposed to improve the identification of environmental risk factors for disease, by applying to epidemiological research, new tools emerging from exposure sciences and high-throughput omic technologies. We have conducted one of the first large scale exposome projects. Environmental contaminants have been argued to lead to the perturbation of one or more “Adverse Outcome Pathways” which consist of a sequence of events including molecular initiating events, biochemical responses, cellular responses, tissue organ responses, individual responses and lastly, population responses. Pathway perturbation can be used as a conceptual framework for organising and evaluating the strength of existing evidence concerning steps necessary for progressing from molecular initiating events to an adverse outcome. Pathway perturbation makes particularly sense within the causal model called the “sufficient-component-cause framework”. This considers exposures or hits that together lead to the outcome under consideration. The model provides a way to account for how multiple factors, whether environmental exposures or genes, combine to result in disease in an individual or population. Another important component of the “pathway perturbation” paradigm is the life-course approach. The early stages of life allow individuals to build-up their ability to respond to strains of different kinds (chemical, physical, biological and psychological) and this response “build-up” constitutes a “reserve” that allows variable resilience, very often depending on socio-economic status in addition to environmental and behavioural agents. These concepts have been systematically developed in the context of the US National Academy of Sciences 21st Century Risk Assessment report (NAS 2017), that refers extensively to the exposome, and have been applied in the research described below.
In Exposomics, we have:
- pooled and integrated information from short-term, experimental human studies and long-term epidemiological cohorts/consortia – including adults, children and newborns – to enable focused investigations to refine environmental exposure assessment based on the concept of life-course epidemiology;

- characterized the exposome, by (a) measuring the external component of the exposome at different critical life stages by employing novel tools and drawing on experience gained in existing EU initiatives (PEM, databases coupled with GIS, remote sensing), with a focus on air and water pollution;

- measured biomarkers of the internal exposome (xenobiotics and metabolites), using omic technologies (adductome, metabolome, transcriptome, epigenome, proteome) in up to 3,000 subjects.

Finally, we have integrated external and internal exposure measures to comprehensively model and assessed exposure to air pollution and water contamination in large population cohorts, through novel statistical modeling.

Overall, Exposomics provides proof-of-principle that an exposome approach can lead to important findings that have an impact both on knowledge of the mechanisms linking exposure to common pollutants with diseases, and on preventive and regulatory action. We highlight in this report several areas of potential application of our findings: low levels of exposure, dose-response relationships, mixtures, biological plausibility of causal associations, and in general a sound contribution to the study of environmentally-induced diseases such as cardio-vascular diseases, asthma, colon cancer and children conditions.

Project Context and Objectives:
There are two broad interpretations of the exposome concept and they are complementary. One, called “top-down”, is mainly interested in identifying new causes of disease by an agnostic approach based on omic technologies, similar to what has been applied in genetics with the GWAS design. This first approach is sometimes called “EWAS”, or “exposome-wide association study”, and utilizes tools such as metabolomics or adductomics to generate new hypotheses on disease etiology. The second general approach is called “bottom-up” and starts with a set of exposures or environmental compartments to determine the pathways or networks by which such exposures lead to disease, i.e. which pathways/networks are perturbed. We have used the latter approach in the Exposomics investigation (Vineis et al, 2016) as is explained below.

It is generally accepted that the majority of important chronic diseases are likely to result from the combination of environmental exposures to chemical and physical stressors and human genetics. There is also evidence that the effects are location-specific and influenced by lifestyle and socioeconomic characteristics. Information on both environmental and genetic causes of disease is growing as a result of large-scale epidemiological research. However, exposure data (including diet, lifestyle, environmental and occupational factors) is often fragmentary (in time and depth), non-standardized, at a crude resolution and often does not include individual level estimates. Additionally, information on environmental factors is often incomplete or inaccurate thereby hindering the accuracy of subsequent estimation of overall risks associated with these factors. As a result, important associations can go undetected. This limitation has recently been framed within the context of the exposome, the environmental counterpart of the genome. The concept of the exposome refers to the totality of environmental exposures from conception onwards, and has been described in detail elsewhere including its external and internal components (Wild, 2005; Rappaport and Smith, 2010; Anon, 2016; Vineis et al., 2009).

The context of EXPOsOMICS is the rapidly developing field of exposure assessment, including the use of omic technologies, according to the concept of the exposome. Historically, a key event was the publication of the report Toxicity Testing in the 21st Century: A Vision and a Strategy by the US National Research Council in 2007. The primary goals of the report were, among others: “(1) to provide broad coverage of chemicals, chemical mixtures, outcomes and life stages, (2) to reduce the cost and time of testing, (3) to develop a more robust scientific basis for assessing health effects of environmental agents” (http://www.nap.edu/catalog/11970/toxicity-testing-in-the-21st-century-a-vision-and-a).
Exposomics aims at partially covering points 1 (mixtures, outcomes, and life stages) and 3.

By comprehensively addressing the integration of the external and the internal exposomes at the individual level, Exposomics provides a holistic and consolidated approach to exposure science. Building upon several EU-funded research projects with rich sets of health data, exposure data, biomarker measurements and publicly available data sources, this multidisciplinary project:

− Pools and integrates information from short-term, experimental human studies and long-term epidemiological cohorts/consortia – including adults, children and newborns – to enable focused investigations to refine environmental exposure assessment based on the concept of life-course epidemiology.

- Characterizes the exposome, by (a) measuring the external component of the exposome at different critical life stages by employing novel tools and drawing on experience gained in existing EU initiatives (PEM, databases coupled with GIS, remote sensing), with a focus on air and water pollution; and (b) measuring biomarkers of the internal exposome (xenobiotics and metabolites), using omic technologies (adductome, metabolome, transcriptome, epigenome, proteome).

− Integrates external and internal exposure measures to comprehensively model and assess exposure to air pollution and water contamination in large population cohorts, through novel statistical modeling.

Together these approaches will lead to the formulation of a new concept of integrated exposure assessment at the individual level, reducing uncertainty, and assessing how these refinements influence disease risk estimates for combined, multiple exposures and selected diseases. The scientific questions to be addressed by the project are:

1. Is it possible to refine exposure assessment to air pollution and water contaminants using a combination of personal exposure monitoring and omic technologies?

2. Will that refinement lead to more accurate estimates of the association with selected diseases, by reducing measurement error?

3. Do new approaches allow the investigation of the effects of mixtures in addition to single components?

4. Do they improve the investigation of dose-response relationships?

5. Is it possible to strengthen causal reasoning by using the “meet-in-the-middle” concept, i.e. investigate the temporal sequence of exposure, biological pathway perturbation and disease onset?

6. Is it possible to use the exposome approach to study the life-course epidemiology of environmental diseases?

This report describes methods and achievements in non-technical language, aimed at translation into the practice of risk assessment. Technical appendices describe methods in greater detail.

Project Results:
1. Filling the knowledge gap

In response to the six general questions above, this report has selected a few research priorities with relevant and practical implications for policy making and stakeholders. Some key questions are:

- Can we consolidate our knowledge on the health effects of two important exposures, air pollution and water contaminants, reinforcing causal assessment?
- Can we detect variation in exposures in a finer way than with the usual tools of epidemiology?
- Can we detect the effects of low and very low levels of exposure using omic biomarkers?
- How can we exploit omic measurements to study mixtures?
- Can we use improved exposure assessment to calibrate estimates of risk and burden of disease?

In the final analysis, this report outlines the methodological aims which include the validation of five sets of omics measured in the same subjects (total number of more than 2,000), and the development of statistical tools to allow the analysis of very complex datasets.

Supplementary materials include: A statistical “toolkit” and definition of the concept of “omics” all of which combine to provide greater clarity of its results.

2. The current legislation and its assumptions

2.1. Air

For PM2.5 the Ambient Air Quality Directive (EU, 2008) introduced a target value (25 μg/m3 annual mean) to be attained by 2010, which became a limit value starting in 2015. In 2014, about 8% of the EU-28 urban population was exposed to PM2.5 above the target value, with large disparities across countries. The WHO AQG recommended level is more stringent with 10 μg/m3 annual mean and accordingly a greater percentage of the urban population is exposed to levels above the World Health Organization (WHO) limit (91% in 2014).

2.2. Water

The WHO has set guideline maximum limit values for each individual TriHaloMethane (THM) and for total THMs. The suggested limit value for chloroform is 300μg/l, for bromoform is 100μg/l, for DBCM is 100μg/l and for BDCM is 60μg/l. Total THMs should not exceed 100μg/l. In the EU the limit value for TTHMs is also set to 100μg/l. However, there is no maximum limit value for the individual THMs. Some European countries (Denmark, Germany, Italy, Hungary, Austria, Netherlands, Norway) have decreased the TTHM limit to even lower levels (25-50μg/l) and few (Slovakia, Spain) have set maximum limit values for some individual THMs. The WHO recommends that concentrations of DBPs should be as low as possible but little is known about the low-dose effects.

Methodology and exposure measures in Exposomics

2.3. General study design

The studies included in Exposomics are represented in Supplementary Figure 3.1.1 and are described in Supplementary methods.

Phase 1 - We have selected and integrated subjects, samples and data from three types of existing studies: Experimental Short-Term Studies (STS), Mother-Child Cohorts (MCO) and Adult Long-Term Studies (ALTS). Collectively, they reflect all life stages from conception to old age. We have measured the external exposome component for air and water contaminants by performing extensive, repeated Personal Exposure Monitoring (PEM). Fresh blood samples were collected from all individuals undergoing PEM, i.e. individuals in STS and those from a representative subsample from MCO and ALTS, selected on the basis of contrasting exposures as estimated by traditional exposure assessment methods. In these samples, plus already stored samples of the recalled subjects from MCO and ALTS, Exposomics has conducted untargeted omic analyses the aim of which was to look for new biomarkers of exposure to chemicals or mixtures and evaluate intra-individual variation of the internal exposome.

Phase 2 - OMIC profiles were also measured in approximately 2000 stored samples from MCO and ALTS, using untargeted analytical methods, with the aim of evaluating them as predictive of risk by examining their association with health effects, and also generating new hypotheses on disease etiology. We combined external and internal exposome data, Land Use Regression models (LUR) and satellite data to calibrate air pollution exposure estimates (e.g. PM2.5 ultra-fine particles and black carbon) obtained using traditional methods in MCO/ALTS and used these refined estimates for risk assessment and burden of disease evaluations. Descriptions of the studies are contained in Supplementary methods.

2.4. Air pollution

Epidemiological studies have traditionally assigned exposure based on monitoring results of a central monitoring site in the city of residence (Jerrett et al., 2005). In the last decade, methods for more individual exposure estimates have been applied, including land use regression models (LUR) and dispersion models (Jerrett et al., 2005; Hoek et al., 2008). A number of studies have shown differences between long-term average personal and ambient air pollution exposure (Van Roosbroeck et al., 2005). In the latter studies, integrated personal samplers were used that provided a single sample over 1–2 days. Hence no assessment of the contribution of different microenvironments to total exposure (and difference) was possible. Recent advances in air pollution monitoring technology mean that personal monitoring can be undertaken with real-time monitors, capable of logging minute-to-minute variability in exposures. Sensors are portable and sensitive. Coupling the new air pollution monitors with GPS and accelerometry from smartphones and information on personal activity patterns, the potential now exists to differentiate between exposure during journeys and fixed-site locations and to assess the contribution of different micro-environments to the magnitude and variability of environmental exposures. At the beginning of Exposomics, no product existed that integrated these technologies into a single personal monitoring system that can be deployed with lay users (e.g. cohorts). This project has addressed this issue. (See Supplementary material 4).

To date, the majority of evidence about the health effects of air pollution from epidemiology and controlled exposure studies is based on particle mass as the measure of exposure. Consequently, regulatory agencies have adopted mass-based ambient air quality standards. Yet particulate matter (PM) is a complex mixture, and particles of different size and composition are likely to have different toxic effects. The EU-funded ESCAPE study had comprehensively characterized sources of outdoor air pollution and developed ambient LUR models for PM10, PM2.5 and NO2. Models are currently in development for elemental composition (XRF; X-ray fluorescence), EC/OC (elemental carbon/organic carbon) and PAHs (polycyclic aromatic hydrocarbons). However, there is a need to develop models for ultrafine particles for which the long-term health effects have been poorly studied because of difficulties in exposure assessment. This is now possible using an innovative mobile monitoring design that has been shown to be reliable and cost-effective in recent studies (Klompmaker et al., 2015; Montagne et al., 2015). One of the properties of particles likely to reflect toxicity is their oxidative potential (OP). By analyzing the spatial and temporal variability of the OP of particulate matter collected on filters, the determinants of that variation are characterized, and new, spatially resolved air pollution models for OP have been developed.

Air pollution models alone, however, only provide information on ambient outdoor pollutant concentrations. Recent advances in GIS (e.g. route modeling) (Gulliver and Briggs, 2005) and micro-environmental models (e.g. indoor-to-outdoor), have led to the development of more detailed, personal exposure models which can be fed by rich data sources on detailed population time-activity patterns. To advance the state of the art Exposomics provided an innovative framework for air pollution external exposome, via the following steps:

1) The scientific partners have integrated instruments from equipment manufacturers (e.g. DiSCmini for UFP and BGI pump for PM2.5) to develop a novel integrated personal monitoring system comprising UFP and PM2.5 personal air monitors with smartphone technology aimed at characterization of micro-environments, activity patterns, and inhalation rates. One of the main technological innovations is the collection of various measurements and the ability to access them via a single source (i.e. smartphone).

2) Deploy the new personal monitoring system among cohort members in a subset of five study areas, covering different sites (e.g. city centre, suburban, industrial, and rural), to collect the largest series of detailed personal exposure measurements of UFP, PM2.5 and activity data in Europe to date (with simultaneously collected blood samples for omics).

3) Develop models in longitudinal studies such as ESCAPE http://www.escapeproject.eu enriched for all the cohorts in the study, and develop methods to transfer these models of air pollution concentrations back in time.

4) Undertake a UFP air pollution “mobile monitoring campaign” in the study areas where PEM was also performed to develop and validate the new land-use regression models for UFP.

5) Apply the Oxidative Potential (OP) depletion analysis technique to extend PM metrics to look at OP. PM2.5 filters collected during fixed-site outdoor monitoring in selected ESCAPE areas have been analysed to detect the spatial and temporal variability in PM2.5-related OP and subsequently create new land-use regression models for OP. We compare below and assess the spatio-temporal differences in exposure estimates between OP with those from traditional particulate metrics (e.g. PM10, PM2.5).

6) Compare PEMs in 3) above with those from residential address locations to quantify the correlation between the two. We further quantify the spatial and temporal micro-environmental contributions to total exposures.

7) Assess the potential for exposure variability (or misclassification) in exposures, e.g. for UFP and PM2.5 where we have both LUR models and measured exposures. Compare traditional (i.e. residential address) and new methods to inform the omic studies (see below) and health risk assessment.

8) Develop a new Europe-wide air pollution model for PM2.5 and methods to apply exposure models for the other pollutants (OP, UFP) to other cohorts and across Europe for health risk assessment. In order to develop exposure models for the pollutants being studied in the external exposome that can be applied to health risk assessment, the transferability of the new LUR models and hybrid models to other countries/regions and time periods is investigated. Develop models which incorporate satellite data either as variables to enhance existing approaches or as a means of calibrating/validating models in areas where routine monitoring data are sparse or do not exist.

9) Provide model estimates for single and multiple exposures (i.e. mixtures) to feed into risk models (e.g. partial least-square regression, ridge regression, Bayesian mixture methods)(Chadeau-Hyam et al., 2013a; Agier et al., 2016) to study the contribution of single compounds and combinations of compounds to adverse health effects in children and adults.

Further details on the PEM technology we have used can be found in the Supplementary materials 4.


2.5. Water contamination

Exposomics also used existing short-term experimental studies (STS) and of long-term population studies (MCO and ALTS) for water contamination. For the short-term scale (PISCINA study) direct measurements of specific chemicals and mixtures in water have mainly been used, while for the life-course studies a combination of models with measurements for validation purposes were utilized. European-based estimates have been obtained further through systematic use of regional and national databases for disinfection by-products (DBPs).

In the PISCINA swimming pool study, external exposure measurements include determination of an expanded range of disinfection by-products (DBP) (trihalomethanes, haloacetic acids, MX, chloramines, haloacetonitriles), overcoming traditional approaches that measure only trihalomethanes. These chemicals have been measured in air, water and/or in biological samples such as exhaled breath (e.g. trihalomethanes) and urine (haloacetic acids) from study subjects. The external exposome is complemented with omic analyses and supported (for phenotypic anchoring) by in vitro assays such as the Salmonella (Ames) mutagenicity test (Maron and Ames, 1983) and mammalian cell chronic and acute cytotoxicity (van Leeuwen et al., 2006). Biological samples (blood, urine, exhaled breath) have been obtained immediately before and after swimming in the pool. Repeated biological samples have been collected after swimming to cover different formation/elimination kinetics of biomarkers, thus contributing to an improved quantitative evaluation of internal exposures.

In the mother-child cohorts, external exposure measurement includes determinations of a range of DBP chemicals in drinking water (trihalomethanes, haloacetic acids, haloacetonitrile). Measurements for some of these chemicals are available from the EU-funded HiWate project (https://www.researchgate.net/publication/ 24037941 Health Impacts of Long-Term Exposure to Disinfection By-Products in Drinking Waterin Europe Hiwate). In the colorectal cancer study (MCC), exposure modeling of DBPs is based on the evaluation of lifetime residential history together with the collection of historical information on DBPs in the relevant regions and water toxicity testing from the short-term studies.
Finally, exposure to specific DBPs has been derived for European populations from routinely collected information in each country. These data, not centrally available in the EU, come from focus contact points in each country (expanding work completed in INTARESE, and HiWate). In addition, a theoretical framework for future routine evaluation of water contaminant information in Europe is proposed below.

2.6. Methodology for omic measurements and annotation

By permitting the simultaneous analysis of large numbers of potential targets without recourse to prior hypotheses, omic technologies provide unique opportunities for the discovery of a new generation of biomarkers of exposure and disease risk, significantly enriched by mechanistic information (Chadeau-Hyam et al.,2011; Lan et al., 2004; Smith et al., 2005; Perera et al., 2009; Zhang et al., 2016a; Zhang et al., 2016b; Bictash et al., 2010; Nicholson and Rantalainen et al., 2011; Wang-Sattler and Yu et al.,2008; Holmes et al., 2008; Calderón-Garcidueñas et al., 2004). The EXPOsOMICS project both advances omic technologies that are related to identification and quantification of environmental exposure (metabolomics, adductomics) and applies multiplex omic technologies that measure downstream effects of environmental exposures (epigenome, transcripts, miRNA, proteins). Omic technologies are deployed in an untargeted manner in both the short-term and the long-term studies, to a total of approximately 3000 samples (Table 3.4.1 in Supplementary methods). A short definition of omics is contained in the “omics” Supplementary appendix.

Metabolomics (MS technology) - Plasma or serum samples have been analysed by high resolution mass spectrometry (MS) coupled to UPLC. Metabolic features characterizing exposed groups are identified by multivariate statistics with appropriate correction for the False Discovery Rate (FDR), and efforts to identify the discriminating metabolites are made. A database of known biomarkers from the main environmental contaminants in air and water as measured in various populations has been compiled from the scientific literature into the Exposome-Explorer database, as one of the foreground products of the project (http://exposome-explorer.iarc.fr/).

Adductomics - Application of adductomics to the exposome concept involves an untargeted investigation of the internal exposome based on the measurement of complete categories of features (e.g. protein adducts). The untargeted approach has been realised by focusing on a specific locus of HSA, and using MS of hydrolysis products to profile covalent modifications over a range of masses.
Intermediate omic biomarkers - Additional omic technologies for the analysis of biological molecules have been used to measure the more downstream effects of environmental exposures. These include the effect on the transcriptome, the epigenome and the proteome. Measurements have been conducted using:

Transcriptomics: (Agilent platform) (number of signals per sample: 44 k)

Epigenomics: Global DNA methylation (Illumina platform) (number of signals per sample: 450 k); microRNA analysis (Agilent platform)

Targeted proteomics: (Luminex Multianalyte Profiling platform for inflammation-related proteins) (number of signals per sample~30). Further technical details are contained in the Supplementary materials 5.

3.5. The “statistical toolkit”

There were several statistical challenges in the work we have undertaken: (a) very large datasets (omics) usually with small numbers of subjects; (b) fragmented nature of the overall database, requiring data integration; (c) unknown nuisance parameters for most omic data (i.e. technical variables that could influence intra- and inter-individual variability); (d) large confidence intervals at low exposure and biomarker levels; (e) complex nature of some study designs, notably Oxford Street (double exposure, three repeated samples, cross-over) and Piscina (pre-post measurements, confounding from physical activity). Other innovative tools have been developed specifically for this project, i.e. (a) models for longitudinal analyses based on the compartmental model; (b) PLS models as a potential instrument for the analysis of mixtures; (c) tools for “cross-omic” analyses, in particular the differential network approach. Statistical tools have been elaborated within a Working Group, jointly between Helix and Exposomics (Agier, L. et al “A systematic comparison of linear regression-based statistical methods to assess exposome-health associations”. Environ Health Prsopect. (2016) 124(12): 1848-1856.)

The technical solutions found to solve the problems listed above are described in the published or submitted papers and summarized in the Supplementary Statistical Toolkit.


Results

3. Air pollution: The gaps filled by Exposomics

3.1. Contribution to the study of microenvironments and intra-individual variability

Exposomics has allowed major advancements in exposure science, including:
- Personal exposure monitoring (PEM) in ~200 individuals in Basel, The Netherlands, Norwich, Turin (adults) and Sabadell (children) of PM2.5 and ultra-fine particle (UFP), allowing insight into micro-environmental (home, travel by mode, work etc.)
contributions to exposures;
- Improved collection and processing of data on individual positioning (GPS) and accelerometry from smartphones (via comparisons with bespoke devices);
- New PM2.5 and NO2 Europe LUR (land use regression) models for harmonised exposure assessment, and reaching out to cohorts without models from previous studies (e.g. ESCAPE);
- Oxidative potential (OP) of PM2.5 and UFP land use regression (LUR) models;
- Detailed analysis of modelled (e.g. UFP LUR) versus measured personal exposures (i.e. misclassification related to traditional methods).

Table 4.1.1 in Supplementary materials shows the data collected in the Personal Exposure Monitoring campaigns (PEM), and Figure 4.1.1 the distribution of UFP by area and site type. Figure 4.1.2 shows the contribution of different micro-environments to UFP exposure in the Dutch PEM, and evidences the role played by home exposures. Figure 4.1.3 in Supplementary materials shows the new Western Europe-wide LUR model for PM2.5 produced within Exposomics and its validation in two datasets (measured vs predicted exposure). A similar model has been produced for NO2.

Ultrafine particles - An original and relevant achievement of Exposomics is the development of a UFP-LUR model, based on the following features:
• Repeated (3) short-term (30-minute) measurements (using the TSI 3007 CPC device) in six areas (Basel, Heraklion, Norwich, The Netherlands, Sabadell, Turin);
• 150-161 (except 240 in The Netherlands) long-term UFP estimates (temporally adjusted from 3*30 measurements);
• Measurements sites split in 10 stratified groups; iteratively 9 groups used to develop a LUR model and the held-out group each time used for model evaluation;
• Combined-areas LUR model developed and validated using the same strategy as above;
• Models used to predict 24h home-outdoor UFP measurements in Basel and The Netherlands;
• Nearby traffic included in all models (population, industry, airports, and restaurants also frequent).

Figure 4.1.4 shows the distribution of UFP levels in the five areas, with higher levels in Sabadell (Spain) and Turin. The model validation was good, with external validation R2 (applying models to the average of 24h UFP measurements in Basel and The Netherlands) of 54% and 49% for area-specific models. Model performance was much stronger when applied to 24h measurements than short-term measurements (average of 30 minutes), i.e. short-term UFP measurements are robust in longer-term exposure prediction.

Oxidative potential - Oxidative potential (OP) of particulate matter (PM) has been proposed as a biologically-relevant exposure metric for studies of air pollution and health. PM2.5 filters were exposed for two weeks in ESCAPE and SAPALDIA projects to collect particulate and test it for oxidative potential. Consumption of ascorbate (AA) and Glutathione (GSH) (two anti-oxidants) in a biological model (synthetic respiratory tract lining fluid ) was calculated for between 2 and 4 filters per site in Basel, Catalonia, London-Oxford, The Netherlands, and Turin; Annual average OPAA and OPGSH was estimated for 20-39 sites depending on the study area (no GSH available for London-Oxford), and OPAA and OPGSH LUR model development for each area and a combined-areas model (using linear mixed effects with random slopes/intercepts on area). Figure 4.1.5 (Supplementary material) shows the distribution of oxidative potential LUR models by study areas. Validation of OP LUR models suggests that they can be used in addition to other air pollutant estimates and their relevance is in their biological significance.

Executive summary:
We have developed Western Europe spatial models that provide harmonised exposure assessment for PM2.5 and NO2, and new outdoor spatial models for UFP and oxidative potential (OP) (with variable performance - poor to good - between areas). Personal Exposure Monitoring (PEM) in five areas among ~200 individuals allowed new insight into the contribution of microenvironments to exposures. Cheap validated sensors were not available at the start of the project. Although recently more sensors have become available all of them still have insufficient accuracy, e.g. in areas with relatively low levels/contrasts in air pollution). However, it is anticipated that these will become available in the near future allowing more coverage in space - and time. Methods developed in Exposomics will allow for comparisons with data provided by such new developments.

3.2. Improvements in exposure assessment and calibration of estimates

The assumption underlying calibration of risk estimates was that the application of correction methods for errors in classification in the Exposomics studies could lead to enhanced sensitivity risk analyses. In Deliverable 9.1 several correction approaches were suggested for implementation in Exposomics: (a) regression calibration, using the measurement data from the PEM studies as validation dataset. This method has been applied by Spiegelman et al. in several different scenarios. (b) post-hoc correction for asymptotic bias (previously applied by Szpiro and Paciorek). (c) computing standard error estimates that account for measurement error. For the methodology and technical details we refer to deliverables 9.1 and 9.2.

As proof-of-principle calculations to derive de-attenuated effects estimates, regression calibration using the measurement data from the PEM studies as validation dataset was applied to PM2.5 and PM2.5 absorbance, in association with ischemic heart disease incidence, asthma incidence and total mortality. Table 4.2.1 in Supplementary materials shows the de-attenuation factors. To exemplify the procedures, we describe them here for study participants of the EPIC-NL cohort investigated for the incidence of ischemic heart disease:
• Ambient PM2.5 (European model) and PM2.5 absorbance estimates (ESCAPE estimates) were available for 31,089 individuals.
• A total of 3,971 CVD events and 2,607 deaths were recorded.
• 39 individuals underwent personal exposure monitoring in Utrecht to calculate Dutch de-attenuation factors
• Cox proportional hazard regression models were produced with age as the time variable.
• Hazard ratios (HRs) were expressed for each 5 μg/m3 increase in PM2.5 and for each 1 x10-5/m3 increase in PM2.5 absorbance.
• Covariates included in the models were: gender, BMI (continuous), marital status (single, married/cohabitating, divorced, widowed), and educational level (primary school, secondary school, university)
• The ‘calibrated’ HR and 95% CI for PM2.5 and PM2.5 absorbance were computed using the method proposed by Spiegelman et al. and implemented using the “Blinplus” macro for SAS
• Outdoor measurements were used as surrogate exposures
• Personal measurements (PEM) were used as “true” exposures.

Results are shown below. All Hazard ratios (HRs) are expressed for each 5 μg/m3 increase in PM2.5 and for each 1 x10-5/m3 increase in PM2.5 absorbance.

Total mortality in the Netherlands
Non-calibrated Calibrated
Hazard ratios (95%CI) Hazard ratios (95%CI)
PM2.5. 1.25 (0.89-1.75) 1.46 (0.81-2.62)
PM2.5. Absorbance 1.22 (1.01-1.46) 1.38 (1.00-1.89)

Incidence of cardiovascular disease in the Netherlands
Non-calibrated Calibrated
Hazard ratios (95%CI) Hazard ratios (95%CI)
PM2.5. 1.12 (0.85-1.48) 1.22 (0.76-1.96)
PM2.5. Absorbance 1.18 (1.02-1.37) 1.31 (1.01-1.69)

An identical procedure was used in Torino with the following results. The correlation between outdoor measures and PEM recorded values is shown in Supplementary Figure 4.2.1.

Total mortality in Torino
Non-calibrated Calibrated
Hazard ratios (95%CI) Hazard ratios (95%CI)
PM2.5. 1.08 (0.87-1.35) 1.25 (0.68-2.31)
PM2.5. Absorbance 1.15 (0.89-1.48) 1.25 (0.83-1.91)

Incidence of cardiovascular disease in Torino
Non-calibrated Calibrated
Hazard ratios (95%CI) Hazard ratios (95%CI)
PM2.5. 1.29 (1.08-1.55) 2.03 (1.20-3.44)
PM2.5. Absorbance 1.08 (0.84-1.37) 1.13 (0.76-1.68)

Adult-onset asthma in Sapaldia cohort
Finally, the same procedure was applied to asthma data in the Sapaldia cohort and results are shown below for PM2.5 as estimated from the Europe-wide model described above. Odds ratios (ORs) are expressed for each 5 μg/m3 increase in PM2.5.
Non-calibrated Calibrated
Odds ratios (95%CI) Odds ratios (95%CI)
PM2.5. 1.15 (0.93-1.42) 1.40 (0.83-2.37)

Executive summary:-
As expected, the use of de-attenuation factors obtained from Personal Exposure Monitoring in comparison with outdoor measurements allowed us to improve risk estimates by reducing exposure classification error. This led to considerably higher estimates of the association between PM2.5 and total mortality, cardiovascular disease and asthma (on average risks are a factor 2 to 3 higher). The new risk estimates thus obtained have been used to re-estimate the burden of disease attributable to air pollution as shownin table 4.2.1 in Supplementary 1.

4. Calibrated burden of disease

As a demonstration, but without estimating the actual uncertainty in the calibrated burden of disease estimates, we applied the new risk estimates described above to the air pollution attributable burden of disease based on the population weighted average PM2.5 exposure as reported in the European Air Quality Report 2016 and burden of disease estimates retrieved from the Global Burden of Disease Database of the Health Metric Institute in Seattle. Table 4.2.1 (Supplementary materials 1) summarizes the burden of disease attributable to PM2.5 exposure of EU member states and Switzerland for all cause deaths excluding injuries above age 30, YLLs for overall deaths above age 30, ischemic heart disease incidence above age 30, and asthma incidence above age 30.

4.3. Biological plausibility: the meet-in-the-middle approach

The simplest way to connect an external exposure with a disease outcome through a set of intermediate biomarkers is the “meet-in-the-middle” approach. This consists of measuring intermediate biomarkers (often with an agnostic omic investigation), and relate them retrospectively to measurements of external exposure and prospectively to a health outcome (disease, or ageing, or other outcomes) in the context of a longitudinal investigation. If the same set of markers is robustly associated with both ends of the exposure-to-disease continuum, this is a validation of a causal hypothesis. This approach also corresponds to a refinement of one of Bradford Hill’s guidelines for causality assessment, i.e. biological plausibility, and is made possible both by the technological developments in omics and external measurements and by the existence of long-term longitudinal population cohorts with biological samples stored for many years. As we show below with the example of coronary heart disease, previous investigations either analyzed biomarkers in relation to exposure to air pollution, or they analyzed biomarkers of disease risk, but did not look at both associations in the same set of individuals with a longitudinal design.

To exemplify how the Exposomics data can be used to perform a meet-in-the-middle analysis, we have considered (a) the mediation exerted by protein and methylation markers on the relationship between air pollution and coronary heart disease; and (b) the mediation exerted by metabolomics in both CVD and asthma. Here we describe the methodology used for (a), while results are reported in different sections below.

Coronary heart diseases (CHDs) are among the major causes of death and disability worldwide. Exposure to ambient air pollution has been linked to a wide range of adverse health effects, including mortality and morbidity due to respiratory diseases and CHD (Wolf et al. 2015). Several epidemiological studies have reported an increased risk of mortality from all-causes, cardiovascular disease and respiratory disease associated with long-term exposure to fine and coarse particulate matters (PM2.5 and PM10), nitrogen oxide (NOx), nitrogen dioxide (NO2) and elemental carbon. Mechanistic explanations have been put forward to explain the effect of air pollution on the cardiovascular system, particularly oxidative stress and inflammation (Uzoigwe et al. 2013; Newby et al. 2015). To assess causality, ideally one needs evidence from a ‘meet-in-the-middle’ (MITM) approach, i.e. one in which biomarkers are associated retrospectively to exposure and prospectively to disease in the same subjects. Briefly, in the CCVD (cerebro-cardiovascular disease) study (Fiorito et al, submitted) study participants were part of the Italian component (Turin and Varese centers) of the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort. From hospital discharge records, we have identified all newly diagnosed cases of CCVD and revascularization that arose in the cohort during 12.2 years of follow-up on average. For biomarker analyses (inflammatory proteins and genome-wide DNAm) and to test the MITM hypothesis a nested case-control study was designed in the cohort using the incident density sampling method including 386 samples (193 matched pairs). Criteria for case-control pair selection and matching based on baseline characteristics were: never smokers and ex-smokers since at least one year, one-to-one matching by smoking (never/time since quitting), gender, age groups (no more than 2.5 years difference at recruitment), season and year of recruitment in the cohort. Exposure estimates are those described above from Exposomics. Figure 4.3.1 (Supp. matrls.1) reports a summary of the timeline for the collection of relevant variables (Figure 1A), and a schematic description of the study hypothesis (Figure 1B). Similar procedures have been used for the MITM in application to asthma, CVD and metabolomics. Results are reported below.

4.4. Identification of early markers at low levels of exposure, and consistency across omics and across studies - How personal measurements help identify OMICS markers of air pollution

Our PEM studies, described above, had different goals including the investigation of microenvironments and the study of the impact of low levels of exposure on the five omics measured. In the PEM we have included two components: a very detailed measurement of short term exposures, including repeated samples, and multiple omics. We have attempted to overcome some limitations in the use of existing cohorts: relatively crude exposure assessment; only one sample per individual; samples often stored for many years; impossibility of assessing acute effects of changes in air pollution levels; links with clinically relevant acute health outcomes often unclear. These limitations were addressed through detailed 24hr exposure assessment, analyses in blood collected following an optimized protocol, two exposure and omic measurements per individual, and blood pressure and lung function measurements collected at the same time as the blood collection. Details on PM2.5 and UFP measurements are given in Supplementary material. The results concerning omics are described here below.

The following protein markers were associated with at least one pollutant: CCL22 - significant positive association with PM2.5 outdoor absorbance, plus indications of positive association with personal PM2.5 and PM2.5 absorbance; G-CSF - significant positive association with personal UFP; CXCL-10 - significant positive association with personal UFP. G-CSF was independently validated in the Piscina-air study: PEM study beta=0.32 p-value=0.02 (FDR 0.15); Piscina beta=0.25 p-value=0.04 (FDR 0.17). However, G-CSF was not associated with modeled exposure to UFP.
In the metabolome-wide analysis, after multiple test correction there are a limited number of significant signals, but statistical analyses are still on-going. The main result was that various exposures (particularly PMC, aka UFP, and PM2.5) were associated with different metabolites as shown in Figure 4.4.1 (Supp. matrls.1).

We also considered the role of repeated measurements. We used individual repeated measurements of UFP in two 24-h campaigns, demonstrating large intra-individual variability. A variable proportion of the total variance for proteins was explained by intra-individual changes, ranging from 2% for EGF to 83% for CRP (24% for G-CSF).

Executive summary: -
This is still a work in progress. We observed some associations with omic markers, particularly proteins, but findings may be enriched by cross-omics analyses. It is currently difficult to establish whether the changes reflect acute or long-term effects and we need to contrast findings from the PEM study to findings from cohort studies. An interesting observation is the lack of overlap between metabolomic signals for single pollutants (UFP vs PM), that may suggest the ability of omics to detect pollutant-specific biological effects (see also the section below on mixtures). The PEM studies can inform cohort studies via regression calibration as explained elsewhere in this report; and by providing insight into marker stability. There are limitations related to low statistical power (low levels of exposure and small omic changes), and multi-city designs are needed for contrast, but these complicate inferences.

4.5. Fingerprints of air pollution via adductomics

One of the main purposes of adductomic analysis is to be able to disentangle the fingerprints left by different exposures. In the first analyses of adductomic datasets (that are still provisional) we have been able to detect some adducts that seem to be associated with smoking, different from others that might be more typical of air pollutants. Figure 4.5.1 (Supp. matrls.1) shows the adducts that overcame the Bonferroni threshold for a comparison between smokers and non-smokers (N=40) (since the current method does not allow annotation of adducts, this feature is indicated as D111.6). Figure 4.5.2 (Supp. matrls.1) shows adducts putatively associated with air pollution in 60 non-smokers. Though annotation is incomplete all of these are different from the ones identified above for smoking.

Another question was whether adductomics could separate different components in a mixture. Applied to Piscina samples (comparison pre vs. post-swimming), the top hits represented in Figure 4.5.3 (Supp. matrls.1) were found for the totality of THM. Exactly the same hits were found for two specific compounds (Bromodichloromethane and Bromoform, on the right), suggesting that the contribution to total adducts might mainly come from these two brominated compounds. The levels of these adducts increased after exposure.

4.6. Is it possible to disentangle different components in a mixture?

Air pollution is a mixture like many other real-life human exposures (tobacco smoke, food, water, etc.) Disentangling the separate effects of components of a mixture has considerable advantages, both for a mechanistic understanding of causal associations, and for practical preventive purposes (to be able for example to eliminate specific components). However, epidemiological studies very often do not allow a clear separation of mixture components, both for limitations in exposure assessment and for high correlation among them (e.g. NO2, NOx and PM in air pollution). One of the assumptions of Exposomics is that improved external exposure assessment and the investigation of omics can improve our knowledge of mixtures since different components may act differently on omic outcomes and pathways (i.e. different pathways may be perturbed). We tested this assumption in the Oxford Street Study (see design above), where we had continuous personal exposure monitoring for PM10-2.5 PM2.5-0.1 ultrafine particles, carbon black, NO2, NOx, temperature and relative humidity; measurements of multiple physiological parameters related to spirometry, exhaled NO, and arterial stiffness; while blood was collected for omics at 3 time points: before the walk, at 2h and 24h post-exposure. There was a good contrast between exposures in Oxford Street vs Hyde Park. (Figure 4.6.1 Supp. matrls.1)

Metabolomics - Agnostics metabolomics originated 5, 77, 13, 6 hits respectively for carbon black (CBLK), NO2, PM10 and PM2.5 at p-value<0.05 after Bonferroni correction. Stability analyses (outliers) and sensitivity analyses did not influence the results. Details of statistical analyses are described in the Supplementary methods, and results refer to a comparison between Oxford Street and Hyde Park and before and after the walk.

Supplementary Figure 4.6.2,(Supp. matrls.1) shows that there is very little overlap between components of the mixture, a result that is in full agreement with the PEM results shown above, but opposite to what was found for THM in water.

The annotation procedure described in Supplementary materials led to the identification of three compounds that all decreased after exposure, 10-nitro-9E-octadecenoic acid; 10-Nitrooleate; and 9-nitrooctadec-9E-enoic acid; and two other compounds that increased, Palmitoyl-L-carnitine and 9-Decenoylcarnitine. Nitro fatty acids are well established anti-inflammatory mediators involved in the resolution of inflammation. L-Palmitoylcarnitine is a long-chain acyl fatty acid derivative ester of carnitine which facilitates the transfer of long-chain fatty acids from cytoplasm into mitochondria during the oxidation of fatty acids; it accumulates in ischemic myocardium and may contribute to myocardial damage. Also 9-Decenoylcarnitine is involved in myocardial infarction (MI), being down regulated in patients with MI. Mitochondrial oxidation of long-chain fatty acids is an important source of energy for the heart and the skeletal muscle during prolonged aerobic work. When a pathway analysis (Mummichog) was conducted (Table 4.6.1 Supp. matrls.1) NO2 was associated with the Carnitine shuttle (that includes L-palmitoylcarnitine, associated in literature with myocardial damage and cardiac dysfunction), while PM10 was associated with prostaglandin formation from arachidonate, both pathways indicating inflammatory processes.

Transcriptomics and miRNA – Agnostic transcriptomics led to a large number of association with PM10 or PM2.5 (>220) but few with NO2 (11); miRNA analysis led to no signal for NO2 and 12 and 6 signals for both PM10 and PM2.5 respectively. Most signals originating from transcriptomics point at alterations in the immune function. (Table 4.6.2 Supp. matrls.1) Figure 4.6.3 shows once again a broad separation between signals related to different components of the mixture.

Executive summary:-
As opposed to Piscina (mixture of THM highly correlated, see below), both metabolomics and transcriptomics show very little overlap between components of the mixture. We can conclude that omic signals are sensitive, since they occur at very low levels of exposure, and are able to disentangle NO2, black carbon and PM. However, a further distinction in terms of annotated features is still unclear, since metabolomics seems to indicate perturbations of inflammatory patterns involved in myocardial infarction, and transcriptomics in the immune function, with an unclear separation between components of the mixture.

5. Water contamination: Gaps filled by Exposomics

Disinfectants (e.g. chlorine) and organic matter give rise to disinfection by-products (DBPs). These were first detected in 1974 and constitute a complex mixture of about 700 compounds, the most common being Trihalomethanes (THM): CHCl3, CHCl2Br, CHClBr2, CHBr3. The maximum acceptable levels in EU are 100 μg/L; in the USA 80 μg/L. Several common DBPs are animal carcinogens at high doses.

Epidemiological studies have suggested an increased risk of bladder cancer in relation to exposure to THM. Supplementary Figure 5.1 shows original work performed in Exposomics to create a map of exposure levels to THM in European countries (see also Deliverable 4.3). It is clear that exposure is very heterogeneous but that THM contamination is common and in some countries at high level. In this report, we will describe the results of the short-term and long-term studies conducted within Exposomics, and also attempt to estimate the burden of disease. Exposure to THM occurs via multiple routes: inhalation in swimmers and during showers (mainly volatile THM), by ingestion (water, coffee, tea, water-based food and beverages) and via dermal absorption (in swimmers or taking baths).

Short term study – The PISCINA-2 study is a pre-post semi-experimental investigation in a single indoor chlorinated pool involving 116 subjects, 18 to 40 years old, non-smokers, non-professional swimmers. We enrolled 4 volunteers per day, 2 days per week, for 30 experimental days (June, September, October, November and December 2013). The subjects neither swam for 1 week nor did any physical activity for 1 day prior to the experiment. They we asked to swim for 40 minutes. Levels of the most common DBPs were measured in exhaled breath before and after swimming, and in the swimming pool water. We evaluated short-term changes in metabolic profiles in response to DBP exposure through swimming in a chlorinated pool. Two blood samples, collected before and 2 hours after swimming, were used for metabolic profiling by liquid chromatography-mass spectrometry. The association between DBP exposures and levels of each metabolite was separately assessed using multivariate normal (MVN) regression. Series of sensitivity analyses and feature annotation were conducted to facilitate results interpretation. A total of 6,471 metabolic features were detected, and all exposures to DBPs were significantly higher after the experiment (Supplementary Figure 5.2) and with specific THM (Figure 5.3 Supp. matrls.1). Figure 5.2 shows a clear cut separation between pre-swimming and post-swimming blood samples for metabolomics features, but this observation may be due to physical activity during swimming. There is evidence that intense physical activity modifies the metabolic and metabolomics profile. MVN models identified 293 metabolic features that were associated with at least one DBP. Uptake of DBPs and physical activity were strongly correlated and mutual adjustment reduced the number of statistically significant associations. From the 293 features, 20 could be identified as associated to DBPs, corresponding to 13 metabolites including compounds in the tryptophan metabolism pathway.

Whole genome gene expression and microRNA analysis were performed using microarray technology on blood samples collected before and 2 hours after swimming for 41 volunteers in the same experiment. Associations between exposure levels and gene expression were assessed using multivariate normal models (MVN), correcting for age, body mass index and sex. P-values were adjusted using the Bonferroni threshold at 5% and pathway analysis was performed using ConsensusPathDB. MVN-models for the individual exposures identified 1,778 genes and 23 microRNAs that were significantly associated with exposure to at least one DBP (Bonferroni 5% level). Of these, 721 genes and 7 microRNAs were associated to exposure to all DBPs. It was not possible to disentangle transcriptome responses to DBP exposure from those related to physical activity due to swimming, as exposure to DBPs during the experiment was co-linear with physical activity (PA). However, eliminating previously detected transcripts for PA showed still a large number of hits that were associated with DBP exposure. The transcriptional and microRNA changes observed after swimming were linked to bladder and colon cancer risks from previous studies. Interestingly, concordant microRNA/mRNA expresssion have been identified for hsa-mir-22-3p and hsa-miR-146a-5p and their targets RCOR1 and TLR4, both related to colon cancer.
Supplementary Figure 5.4 shows the transcriptomic signals found in association with bromoform. A large number of transcripts and miRNA were in fact associated with most THMs but particularly with brominated ones. Figure 5.5 (Supp. matrls.1) shows the complex network of interactions across mRNAs and miRNAs for this class of compounds. Whether or not the effect of the mixture is greater than the sum of separate effects is unclear and is discussed below.

Long-term study: colorectal cancer – We have conducted a multi-site case-control study (MCC) in 10 Spanish municipalities, including 6105 incident cases (of which 2200 were colon cancer cases) and 3898 population controls selected from rosters of Primary Health Care centres. The response rates in controls ranged between 55% and 80%. Blood samples were collected and five omics were measured according to the Exposomics protocol. Modeling of DBPs was based on the evaluation of lifetime residential history together with the collection of historical information on DBPs in the relevant regions and water toxicity testing from the short-term studies. First, we considered the association with THM markers in blood. Figure 5.6 (Supp. matrls.1) shows that some proteomic markers (mainly interleukins) were detected in association with long-term THM exposure from the MCC study controls. Figure 5.7 (Supp. matrls.1) shows the association between genome-wide methylation and water contamination estimates: THM (top) and nitrates (bottom).

Is there an overall effect of the mixture that is greater than the sum of components?

A second question about mixtures (in addition to the separation of the effects of components) is whether the combined effect is greater than the sum of their components, i.e. if there is an interaction among components. We have developed an original statistical approach that allows analyze of multiple exposures and multiple outcomes. For the sake of simplicity we have used only a limited set of inflammatory proteins as outcome, while exposures are four THM (bromodichloromethane, bromoform, chloroform, and dibromochloromethane) in the Piscina study. We have used a partial least square (PLS) approach to investigate the multivariate inflammatory response to a multivariate set of water contaminants, which is also able to accommodate repeated measurements.

For all exposures a strong contrast was observed before and after the swimming session (p-values < 4x10-16). Strong correlations among exposures (especially for chlorinated compounds in exhaled air) were also observed, particularly in the post-swimming samples. Given such strong correlation, inferring whether the totality of exposures was greater than their sum in terms of effects on the protein markers was difficult. The exclusion of any single exposure or set of them only marginally reduced the variance of proteins explained by components of exposure. This suggests that due to the strong correlation among exposures, the loss of information induced by the exclusion of single compounds could mostly be recovered by other compounds. However, changes in explained variance were more limited upon exclusion of Bromoform, hence suggesting a possible different behaviour of Bromoform in comparison to other di-bromo derivatives and chloroform.

Executive summary:-
The toxicity of some DBPs is proven at high doses in vitro, though toxicity at real life levels has not been well examined. We have provided new knowledge on mechanisms from short and long-term exposure studies in humans. Our study of metabolomics suggests numerous metabolic changes due to swimming in a chlorinated pool. Molecular characterization of these metabolic features allows us to identify those that may be related to exposure changes related to the swimming experiment. Concerning transcriptomics, eliminating previously detected transcripts for physical activity showed still a large number of hits that were associated with DBP exposure. The transcriptional and microRNA changes observed after swimming were linked to bladder and colon cancer risk from previous studies. Epidemiological studies are mostly convincing for an association of THM with bladder cancer. Exposomics has added evidence for colorectal cancer. A Burden of Disease exercise has never been done for water contamination by DBP in the EU or globally and we provide a first attempt. Concerning, the analysis of DBPs as a mixture, despite strong correlations among the measured exposures, drops in explained variance were more limited upon exclusion of bromoform, suggesting a somewhat different contribution in comparison to other compounds. A tentative conclusion is therefore that the overall effect of the four compounds is greater than their sum and is exerted in particular on 8 proteins.

6. Contribution to understanding asthma

A few studies have investigated the association of air pollution with adult-onset asthma prevalence, with significant associations reported either among all participants, never-smokers or females only, though results overall tend to be inconsistent (Jacquemin B et al, Epidemiology 2009; Kunzli N et al. Thorax 2009). The ESCAPE study found an association of incident asthma with PM2.5 that was not statistically significant (HR=1.04 (95C CI 0.88-1.23 for 5 µg/m3 increase in PM2.5) (Jacquemin et al. Environ Health Perspectives 2015).

The following is a description of the results of our MITM approach which aims to identify potential molecular changes lying between the exposure and the outcome which, when intervened upon, will block (some or all of) the causal pathway between exposure and outcome.

A nested case-control study was designed in the Sapaldia cohort in Switzerland. We included self-reported cases of asthma at most recent survey, irrespective of history of asthma and related symptoms. Controls were subjects without any of the following at any of the three consecutive surveys: self-report of asthma; physician diagnosis of asthma; asthma attack last 12 months; asthma medication currently; wheezing without cold; combination of at least 3 symptoms last 12 months (breathless while wheezing; woken up with a feeling of chest tightness; attack of shortness of breath after exercise; attack of shortness of breath while at rest; woken up by attack of shortness of breath). All subjects were non-smokers for a least 1 year before blood was drawn. Omic measures were performed in 204 cases and 201 controls, all in wave 3 (transcriptomics, metabolomics, proteomics, epigenomics and adductomics).

A logistic regression of adult-onset asthma, defined as self-reported age of asthma onset 16 years or older, was conducted with several air pollution metrics, after adjustment for age, sex, education level, BMI, and study area, using the entire never smokers in SAPALDIA cohort, with the following findings:
Air Pollution Metric Odds Ratio [95% CI]
PM2.5 1.09 [0.66 1.79] - per 5 μg/m3 increase
UFP 1.44 [1.09 1.91] - per 5000 particles/cm3
LDSA 1.38 [1.07 1.78] - per 10 μm2/cm3


Metabolome-wide investigation (MWAS)

Metabolites were declared significantly associated with adult-onset asthma if the p-value was < 0.05 after Benjamini-Hochberg correction. None was found, while 237 metabolomic features were found associated with PM2.5. Then we conducted a pathway enrichment based on the Mummichog algorithm (see methods). The latter is a computational algorithm used to predict functional activity directly from non-targeted MWAS results, without direct metabolite identification. We also compared the pathway enriched in relation to air pollutants in both the Sapaldia study and in EPIC-Italy. Pathways that resulted enriched in both investigations were: for PM2.5 Linoleate metabolism and Fatty acid activation; for UFP Linoleate metabolism, Glycerophospholipid metabolism and Glycosphingolipid metabolism. Among these three pathways, Linoleate metabolism and Glycerophospholipid metabolism were also found enriched for adult-onset asthma, being thus confirmed as a MITM pathway. Further chemical identification in the IARC laboratory confirmed the identity of Linoleate, which contributed to enrichment of both MITM pathways. Linoleate in vitro regulates IL8 (proinflammatory cytokine). Linoleate consumption was positively associated with eczema in Japanese children, and atopic children had higher level linoleate and lower levels of its metabolites. Lipoxygenases, enzymes metabolizing linoleates, have been suggested as relevant to asthma. Therefore, biological plausibility is high for this MITM signal that confers further credibility to the association between air pollution and asthma onset.

Executive summary:-
Results from molecular mediation are consistent with air pollution impacting on both asthma and CVD (see also next paragraph) via pro-inflammatory and oxidative stress pathways, albeit different molecules may be involved in the two groups of diseases. This is consistent with the accumulation of oxidative molecular damage over years of exposure. Linoleate metabolism was identified as a MITM pathway. Linoleate in vitro regulates IL8 (proinflammatory cytokine) and Lipoxygenases, enzymes metabolizing linoleates, have been suggested as relevant to asthma. Therefore this MITM signal lends causal credibility to the association between air pollution and asthma onset.

7. Contribution to understanding cardiovascular diseases

7.1 Omics and CVD

Air pollution has been associated with a broad range of adverse health effects, including mortality and morbidity due to cardio- and cerebrovascular diseases (CCVD), but the molecular mechanisms involved are not entirely understood. In Exposomics we aimed to describe the molecular pathways involved in the association of CCVD with air pollution focusing on the a priori hypothesis of an involvement of oxidative damage and inflammation, using protein and methylation measurements. Within a cohort of 18,982 individuals from the EPIC Italy study, we designed a case-control study on incident CCVD (193 case-control pairs). We measured air pollution, inflammatory biomarkers, and whole-genome DNA methylation in blood. We have investigated the association of biomarkers with air pollution retrospectively and CCVD risk prospectively according to the ‘meet-in-the-middle’ longitudinal design. In the overall cohort study (948 CCVD cases), exposure to air pollution was significantly associated with increased CCVD risk. Two DNA methylation inflammatory pathways, ‘ROS/Glutathione/Cytotoxic granules’ and ‘Cytokine signaling’, were associated with both exposure to air pollution (p-value for enrichment adjusted for multiple comparisons ranging from 0.03 and 0.05 depending on pollutant) and CCVD risk (p=0.04 and p=0.004 respectively). Also, Interleukin-17 was associated with NO2 (p=0.0004) NOx (p=0.0005) and CCVD risk (OR=1.79; CI 1.04-3.11; p=0.04).

An MWAS was also conducted for CCVD using the same criteria and design as described above for asthma. Concerning the association with air pollutants, we selected the Mummichog pathways that appeared statistically significantly associated in both studies (Sapaldia and EPIC)(see previous paragraph). In addition to PM2.5 and UFP as discussed above, we also found pathways associated with NO2: Pyrimidine metabolism and the Carnitine shuttle (the latter confirmed by further laboratory investigation at IARC).

7.2 CVD, adult onset asthma and UFP: novel finding

An important finding to emerge from Exposomics is the first demonstration of an excess of CVD in association with UFP exposure. This novel result has been facilitated by the exposure assessment tools developed in the project and is relevant in the present historical context because of regulation proposals of UFP in several Western countries. In fact, there is growing evidence that exposure to ultrafine particles (UFP – particles smaller than 100nm) may play an under-explored role in the aetiology of several illnesses, including cardiovascular disease. Using a prospective cohort of 40,011 Dutch residents, we studied the association between long-term exposure to UFP and incident cardiovascular disease using Cox proportional hazard models. These hazard ratios (HR) were compared to HRs for more routinely monitored air pollutants including PM10, PM2.5 and NO2. Long-term exposure to UFP was associated with an increased risk for all incident cardiovascular disease (HR: 1.18 per 10,000 particles/cm3, 95% CI: 1.03:1.35) myocardial infarction (HR: 1.33 95% CI: 0.98:1.80) and heart failure (HR: 1.80 95% CI 1.17:2.77). Similar findings were observed for NO2 and coarse PM (PM between 10 and 2.5 µm) but no such risk was identified for PM2.5. Two-pollutant models identified that the elevated risk estimate remained stable for UFP, while results for PMCoarse and NO2 attenuated to the null. These findings strengthen the evidence that UFP exposure plays an important role in cardiovascular health and that risks of ambient air pollution may have been underestimated based on other air pollution metrics such as PM2.5. Including UFP exposure in future air quality guidelines may result in significant health gains and therefore should be strongly considered by policy makers (Downward et al, submitted).

The finding on the association between UFP and adult-onset asthma in SAPALDIA, which is based on UFP measurement and modeling conducted in the SAPALDIA study areas in a manner equivalent to the ones conducted in Exposomics, is also novel (see above: OR=1.44 [1.09 1.91] per 5000 particles/cm3 ). In parallel to the findings for CVD it reinforces the need for policy to consider regulation of this pollutant in the future.

Executive summary:-
Chronic air pollution exposure can cause oxidative stress, that in turn activates a cascade of inflammatory responses mainly involving the ‘Cytokine signaling’ pathway, leading to increased risk of CCVD; this suggests promising biomarkers for CCVD prevention. Metabolomic analyses confirmed the role of Linoleate metabolism in CCVD and added the Carnitine shuttle as an additional player. Our findings on UFP strengthen the evidence that UFP exposure plays an important role in cardiovascular health and that risks of ambient air pollution may have been underestimated based on other air pollution metrics such as PM2.5.

8. Contribution to understanding child health

8.1 Determinants of birth-weight

Birth-weight is an important indicator of maternal and fetal health and a predictor of many diseases in later life. While birth-weight is known to be strongly influenced by the maternal environment, the determinants of variance in birth-weight are still poorly understood. In a ‘top-down’ exposome approach, we applied untargeted metabolomics of cord blood samples to investigate the biological pathways important in determining birth-weight, which may be perturbed by environmental exposures. The design of the children studies in Exposomics is described in the Supplementary material.

Metabolic signatures were acquired using high-resolution mass spectrometry coupled to liquid chromatography from cord blood samples collected on delivery in the ENVIRONAGE (Belgium), INMA (Spain), Piccolipiu (Italy) and Rhea (Greece) birth cohorts. We performed a metabolome-wide association study for birth-weight on 481 samples, adjusting for gestational age, sex, cohort, maternal height, maternal weight, and paternal height. Laboratory annotation of identified metabolites was conducted through reference to authentic standards. Mediation by the metabolome of factors associated with birth-weight was explored in an attenuation analysis.
We identified 68 metabolites significantly associated with birth-weight after controlling the false discovery rate at 5% (Supplementary figure 8.1). These metabolites included vitamin A, progesterone, docosahexaenoic acid, indolelactic acid, and multiple acylcarnitines and phosphatidylcholines. We observed significant enrichment (p < 0.05) of the tryptophan metabolism, prostaglandin formation, C21-steroid hormone signaling, carnitine shuttle and glycerophospholipid metabolism pathways. Perturbation of acylcarnitine and tryptophan metabolite levels, in particular indicate the crucial role of mitochondrial function in fetal growth. Mitochondria are particularly susceptible to environmental toxicants, and therefore may play a key role in linking the maternal environment during pregnancy to the weight of the baby at birth. Furthermore, Vitamin A was positively associated with both maternal smoking and birth-weight. The observed attenuation of the association between smoking and birth-weight upon adjustment for vitamin A levels, suggests that depletion of vitamin A availability may present a novel causal pathway linking smoking and impaired fetal development. Associations were found between birth-weight, metabolic features and air pollutants, but none were statistically significant and this work is still in progress.These results provide insight into mechanisms that affect birth-weight and may have implications for the developmental origins of diseases in adulthood.

8.2 Air pollution and omics in children

Exposures in utero are relevant not only to early life effects (such as cognitive development or growth) but also to adult health, according to Barker’s developmental origin of adult disease hypothesis (i.e. physiological adaptations during pregnancy in response to changes in environment prepare for post-natal life). A summary of the two most relevant findings coming from the investigation of OMIC changes after exposure to air pollutants are presented, followed by the newly identified molecular signatures of in utero exposure to air pollution across OMICs. Supplementary Figure 8.2.1 shows the distribution of exposure levels in the different children cohorts, with a gradient of particulate matter exposure from ENVIRONAGE (Belgium, rural areas) to Torino (Italy, industrial city).

For the analysis of transcriptome data, a targeted approach is used based on previously identified 34 blood pressure transcripts. (Huan et al. 2015). Of 21 transcripts associated with SBP in adults, 3 were associated with prenatal exposure to particulate air pollution in boys in Exposomics - (FGFBP2, GPR56, PRF1), after Bonferroni correction, but not in girls. Also, particulate exposure was associated with changes in cord insulin and with down regulation of SETD8, a gene involved in mitochondrial function and insulin metabolism.

Supplementary Figure 8.2.2 shows the results of metabolomic-wide (mass spectrometry) analyses in relation to particulates. A limited number of significant features were found, in particular in the cohort with highest exposure (Torino). These differences suggest a dose response and possibly a threshold effect in the metabolome response to particulate matter. Supplementary Figure 8.2.3a shows the overall associations of omics with PM2.5 exposure during pregnancy in the four child cohorts (INMA, Rhea, ENIVRONAGE and Piccolipiu) with cross-OMICs data available. Only a limited number of features were statistically significant at the omics-wide level. We then explored cross-OMIC correlations between the strongest associations with PM2.5 in each OMIC platform, to examine whether the signals reported by each platform related to similar biological processes (Supplementary Figure 8.2.3b). Nine features had significantly significant cross-OMIC correlations, including among others between C-reactive protein and a transcript from the NADH Dehydrogenase gene. Further work to integrate multi-OMIC signals, through multi-variate models is ongoing. The distinct distribution of pollutants in the regions included in the children’s study has proved challenging and has reduced the power of pooled analyses. We are developing the use of non-parametric models to address this issue.

Early life epigenetic programming influences adult health outcomes. Moreover, DNA methylation is known to be sensitive to environmental stimuli and changes during the lifetime. Longitudinal methylome changes were therefore further addressed in the ALSPAC cohort (UK). An integrated analysis of DNA methylation at birth, childhood and adolescence in relation to PM10 exposure was performed and three gene pathways were significantly associated with exposure to PM10 according to KEGG and REACTOME: the GABAergic synapse, the NOTCH1 and p53 signalling pathway. The first has been associated to neurological impairment, the second plays a major role in embryonic development and the third is a well-known cancer pathway.

Executive summary:-
At birth: we found different responses in boys and girls for transcriptomic features in relation to air pollutants. Known candidate transcriptome profiles of blood pressure/ insulin in adulthood were associated with prenatal PM exposure at birth. Top significant newborn PM transcripts are likely to have consequences based on associations with cord metabolites and protein targets. From birth to adolescence: longitudinal early life air pollution measures were associated with alterations in genes involved in neuro-transmission and tumor suppression pathways. These findings are overall relevant to Barker’s hypothesis of early life origin of adult disease and suggest that preventive interventions in early life my pay greater dividends than interventions in late life (Supplementary Figure 8.3).


Potential Impact:
9. Potential Impact

In Exposomics we have extensively applied methods derived from the concept of “exposome”, i.e. a systematic investigation of external exposures at different ages in the life-course, of internal markers with the use of omic technologies, and of the impact on health outcomes. We have focused on air pollutants and water contaminants, by analyzing both birth as well as adult cohorts including experimental studies in humans.

Several of our findings can have an important impact on both future research and policies. We outline some of them in the following:

The main impacts foreseen in the original plan were: (a) the “reduction of uncertainty” paradigm; (b) the role of air pollution and water contamination in disease risk, on the basis of improved exposure assessment; (c) novel chemical risks identified via untargeted omics (“hazard identification”); and (d) the methodological exercise performed on the burden of disease (how the latter changes with improved exposure assessment and risk estimation, for selected exposures and diseases). In addition, we aimed to evaluate whether the parameters listed above have a different impact on adults or children, on different types of diseases and between air and water pollutants. All these aspects are highly relevant for a number of stakeholders, in particular for policymakers and regional and international agencies.

New exposure models and reduction of uncertainty

We have developed Western Europe spatial models that provide harmonised exposure assessment for PM2.5 and NO2, and new outdoor spatial models for ultrafine particles (UFP) and oxidative potential (OP) (with variable performance - poor to good - between areas). Personal Exposure Monitoring (PEM) in five areas among ~200 individuals allowed new insight into the contribution of microenvironments to exposures. Cheap validated sensors were not available at the start of the project. Although more sensors have recently become available, all of them still have insufficient accuracy (e.g. in areas with relatively low levels/contrasts in air pollution). However, it is anticipated that these will become accurate and available in the near future allowing more coverage in space and time. Methods developed in Exposomics will allow for comparisons with data provided by such new developments.

PEM was used to improve our knowledge of exposure to air pollution, including the role of different micro-environments and the degree of classification error associated with previous, more traditional measurements such as Land Use Regression (LUR) Models. As expected, the use of de-attenuation factors obtained from PEM in comparison with outdoor measurements allowed us to improve risk estimates by reducing exposure classification errors. This resulted in considerably higher estimates of the association between PM2.5 and total mortality, cardiovascular disease and asthma (on average risks are a factor 2 to 3 higher). These new risk estimates have been used to re-estimate the burden of disease attributable to air pollution.

The relationship between the external exposome and the internal exposome (omics) is still a work in progress. We observed associations of PEM with omic markers, particularly proteins, but findings may be enriched by cross-omics analyses. It is currently difficult to establish whether the changes reflect acute or long-term effects and we need to contrast our findings from the PEM study to findings from cohort studies. The PEM studies can inform cohort studies via regression calibration as explained above; and by providing insight into marker stability. There are limitations related to low statistical power (low levels of exposure and small omic changes), and multi-city designs are needed for contrast. As opposed to the experimental study in swimmers (Piscina) (highly correlated mixture of water contaminants), both metabolomics and transcriptomics show very little overlap between components of the air pollution mixture. We can conclude that omic signals are sensitive, since they occur at very low levels of exposure, and are able to disentangle NO2, black carbon and PM. However, a further distinction in terms of annotated features is still unclear, since metabolomics seems to indicate perturbations of inflammatory patterns involved in myocardial infarction, and transcriptomics in the immune function, with an unclear separation between components of the mixture.

Concerning water contamination, the toxicity of some disinfection by-products (DBPs) is proven at high doses in vitro, though toxicity at real life levels has not been well examined. We have provided new knowledge on mechanisms from short and long-term exposure studies in humans. Our study of metabolomics suggests numerous metabolic changes due to swimming in a chlorinated pool. Molecular characterization of these metabolic features allows us to identify those that may be related to exposure changes linked to the swimming experiment. Concerning transcriptomics, eliminating previously detected transcripts for physical activity showed still a large number of hits that are associated with DBP exposure. The transcriptional and microRNA changes observed after swimming were linked to bladder and colon cancer risk from previous studies. Epidemiological studies are mostly convincing for an association of THM with bladder cancer, while Exposomics has added evidence for colorectal cancer. A Burden of Disease exercise has never been done for water contamination by DBP in the EU or globally and we have provide a first attempt.

Hazard identification - New technologies for environmental health research – adductomics

Adductomics, the untargeted detection of DNA or protein adducts of endogenous or exogenous origins, is a new field in Exposome research. Most of the studies in this area focused on the untargeted analysis of protein adducts in human serum albumin (Cys34 - the major site of modification) through a method developed in 2013. To date, the majority of studies have been concerned with quality control and methodological development and validation, but the first analyses in some of the Exposomics studies (PISCINA2, PEM, Oxford Street 2) and other epidemiologic studies (ESCAPE/EPIC) are starting to yield promising results with the identification of correlations between specific adducts and different environmental exposures or disease states. Further methodological developments are still required, including the development of new analytical methods and the creation of adduct libraries for annotation, but adductomics is set to become a key component in the study of the Exposome.



Contribution to understanding disease risk

Results obtained from molecular mediation are consistent with air pollution having an impact on both asthma and CVD via pro-inflammatory and oxidative stress pathways, albeit different molecules may be involved in the two groups of diseases. This is consistent with the accumulation of oxidative molecular damage over years of exposure. Linoleate metabolism was identified as a meet-in-the-middle (MITM) pathway. Linoleate in vitro regulates IL8 (proinflammatory cytokine) and Lipoxygenases, enzymes metabolizing linoleates, have been suggested as relevant to asthma. Therefore this MITM signal lends causal credibility to the association between air pollution and asthma onset. Chronic air pollution exposure can cause oxidative stress, which in turn activates a cascade of inflammatory responses mainly involving the ‘Cytokine signaling’ pathway, leading to increased risk of cardio and cerebro-vascular disease (CCVD); this suggests promising biomarkers for CCVD prevention. Metabolomic analyses confirmed the role of Linoleate metabolism in CCVD and added the Carnitine shuttle as an additional player. Our findings on UFP strengthen the evidence that ultrafine particle (UFP) exposure plays an important role in cardiovascular health and that risks of ambient air pollution may have been underestimated based on other air pollution metrics such as PM2.5.

We detected several relevant omic signals in our children studies. At birth we found different responses in boys and girls for transcriptomic features in relation to air pollutants. Known candidate transcriptome profiles of blood pressure/ insulin in adulthood were associated with prenatal PM exposure at birth. Top significant newborn PM transcripts are likely to have consequences based on associations with cord metabolites and protein targets. From birth to adolescence longitudinal early life air pollution measures were associated with alterations in genes involved in neurotransmission and tumour suppression pathways. These findings are overall relevant to Barker’s hypothesis of early life origin of adult disease and suggest that preventive interventions in early life my pay greater dividends than interventions later in life.

Statistics in Exposome Research

There were several challenges in the analysis of large, complex datasets produced by untargeted omics analyses in studies of the Exposome (e.g. simultaneous testing of multiple hypotheses, consideration of multiple correlated exposures, of exposure interactions and non-linear exposure-response relations, and of temporal factors in exposures) together with the statistical tools being developed to address them. Examples of the application of established statistical methods to analyses of data from Exposomics are presented separately, including: a comparison of several multivariate regression-based methods for identifying true exposure-outcome associations from a large set of correlated exposures; the use of techniques for analysing multivariate data, such as multi-level partial least squares (PLS) methods, in the analysis of multivariate exposures from the PISCINA study, for identifying specific molecular signatures representative of entire exposure sets; the use of network representation methods for the identification of key signals or combinations of signals in omics data that play a pivotal role in describing the association between exposure and effect, illustrated by analyses of transcriptomic profiles in the PISCINA study and of epigenome-wide association studies of smoking and lung cancer risk. Finally, the dissemination and training activities in statistical analysis of omics data integrated in Exposomics have been considerably developed and successful.

Impact on stakeholders

The final meeting of the Exposomics project “EXPOsOMICS Final Policy Workshop and Stakeholder Consultation” took place in Brussels, Belgium on 28-29 March 2017. The meeting programme was structured around the main research topics in Exposomics and presentations fell into four main themes: main results and the advances in Exposome research achieved through the project; presentations on other parallel research initiatives on the study of the Exposome in Europe and the USA and their complementarity to EXPOsOMICS; lessons learned from these early studies on the Exposome and how they will shape the future of research on environmental exposure assessment; and finally, the broader implications of Exposome research for hazard identification, risk assessment and policy development on environmental exposures. Additionally, the programme included three plenary sessions - on the External Exposome, the Internal Exposome and on Policy Translation – each of which were led by two discussants who introduced the themes and promoted discussion amongst the participants. This paper summarizes the presentations and discussions at this meeting.

The plenary session on the “External Exposome” discussed the following questions:

• What can be the contribution of exposure science to hazard identification and risk assessment?
• What is the state of the art of new exposure measurement technologies?
• What are the research needs?

The development of mobile technologies and of smaller, cheaper, accurate sensors allows for easier and more frequent collection of data on exposures in research participants (ie. location, air pollutants, noise, diet) and the modelling of exposures in the wider population to numerous environmental factors. Commercial personal monitoring devices will continue to get cheaper and more accurate and in a few years may be accurate enough for research purposes. Commercial apps to automatically track location, mapping of routes and physical activity are becoming more sophisticated and popular and data could be harnessed for research. There is the possibility of increasing accuracy by combining measurements from many sensors, assuming random error. There is also the potential for crowdsourcing data and the use of data from commercial providers, including the use of data from social media and smart homes, as the much larger sample size would aid to reduce the risk of selection bias. Mapping of participant residential information may also be useful in the identification of potential selection biases that otherwise may be difficult to detect. With better understanding of the correlation structure of many exposures, we may be able to measure smaller numbers of exposures representative of a broader exposure profile.

However, there remain a number of methodological challenges. There are currently many poor quality sensors on the market and a need for thorough validation of new and existing ones. It would be useful to provide information on the quality of available devices to the public such as the EPA air sensors toolbox (www.epa.gov/air-sensor-toolbox) that provides general guidance to the public, researchers, and developers on available air pollution monitoring devices. It would also be useful to compare information obtained using different mobile applications such as those developed specifically for research purposes vs. commercially available apps (ie. ExpoApp developed for the European CITI-SENSE project (http://www.citi-sense.eu/) vs. Moves (https://moves-app.com/). Although photo-based methods and wearable cameras have been implemented in epidemiological studies to better understand participant diet or time-activity behavior patterns, further work is needed in the processing of such data to go beyond estimating food volume only to recognising specific types of food or food cooking method. There is a need to define the optimal vs. sufficient level of resolution that is needed for research and policy-making as well as participant adherence and measurement in large-scale populations. There are also outstanding questions regarding the collection, storage, and access to data. Further work to validate data on External Exposome measures using internal measures may also be useful (below).

The plenary session on the “Internal Exposome” discussed the following questions:
• What can be the contribution of omic measurements in hazard identification and risk assessment?
• What are the current limitations?
• What are the most urgent needs in the field of OMICS research?

At present OMICS approaches are useful in biomarker discovery and research, and also for hypothesis generation, but it will take time to validate OMICS approaches to the point that there is sufficient confidence for its use in regulatory and policy decisions. There are both agnostic and targeted approaches with different advantages (hypothesis generation vs. refining knowledge on mechanisms) and the opportunity for cross-validation and discovery of new biomarkers/mechanisms by combining both approaches. OMICS data may also be useful to inform physiologically-based pharmacokinetic modelling (PBPK) to improve internal exposure estimates integrating epidemiology and toxicology approaches or as a tool to develop more cost effective interventions by focusing on the most relevant pathways for prevention.

Limitations of OMICS are different depending on intended use, ie. quantifying exposures or downstream markers of pathways associated with disease outcomes. There is also a limited understanding of biological pathways, particularly interactions between different pathways. Although OMICS have contributed to advancing the field i.e. use of overlaps across different OMICS to identify the most robust findings, there is a need to recognise the danger of over-interpreting results. There is a limited capacity to look at historical exposures - adductomics and methylomics may provide historical data but this still requires validation. There is difficulty in distinguishing effects of exposures from the effects of disease processes associated with them in analyses of the Internal Exposome. There is also a large variation in bioinformatics analyses of OMICS data and consequently a need for increased standardization and reproducibility. There is a need for validation of OMICS approaches (both technical validation and biological validation) and a lack of a platform for data sharing - consider an international initiative to promote data sharing and setting of standards for the reporting and validation of OMICS markers. Finally although the cost of OMICS analysis is decreasing, it remains a limiting factor in most studies.

The plenary session on “Policy translation” discussed the following questions:
• In the light of the philosophy expressed in the US National Academy of Sciences (NAS) report on pathway perturbation, what is the potential contribution of the Exposome paradigm?
• How does it fit into the strategies of environmental and public health agencies, NGOs, regulatory agencies, industry and academy? What institutional actors are necessary?
• How should research be funded to meet the next challenges of Exposome research?

Before discussing the potential contribution of the Exposome to policy development it is important to consider what is meant by policy: guidance, recommendations, laws all have different requirement levels for evidence. It is important to bear in mind the many factors that affect the translation of science into policy, including public and media pressure, economic interests, and political agendas, as examples.

On a basic level, Exposome research can be seen as replicating the approaches of classic risk assessment with more accuracy. This includes improved exposure assessment with the ability to capture correlated co-exposures, complex mixtures, and synergies, the provision of dose-response data including at low-dose exposures, and biological plausibility of exposure-disease associations by bridging experimental and human data. Identifying susceptible sub-groups and critical windows of exposure, monitoring prevalence and level of exposure and evaluating interventions through short-term endpoints and/or mechanism-based markers can also be performed.

However, pathway perturbation is a change of paradigm, a new way of thinking about risk assessment; using pathway analysis to link multifactorial causality with risk decisions. The possibility of evaluating complex mixtures and synergies between compounds is also a change of paradigm from evaluating risk for individual agents. Our understanding of the dynamic changes and interactions in pathways and the way they relate to exposures are still patchy; this limits the way pathway analysis can be used to identify multifactorial aetiology underlying disease, at least at present. Further, current regulatory standards and policy currently focus on animal models for mechanistic evidence and biological plausibility and epidemiological evidence for strength of association as pre-requisites (i.e. Bradford-Hill causality assessment criteria) which are also not yet adapted to the use of Exposomics evidence (pathway analysis/pathway perturbation) for risk assessment.

There is an opportunity for Exposomics to contribute to breaking the institutional silos in policy-making organisations, by promoting integrated approaches that examine the effects of multiple categories of agents in a more holistic approach to risk assessment. However, policy development is typically slow due to powerful pressures from existing interests; this will condition the speed with which novel approaches and data from Exposomics will be accepted for translation into policy. In the face of resistance from external interests, translating evidence into policy requires strong, well organised advocacy; including the consideration of engaging with other groups in society with an interest in the protection of public health and the environment either as a direct or co-benefit. There is also a big gap in the way questions are framed in a scientific vs. a regulatory/policy context and a need to consider from the stage of design how study results can be relevant to and presented in a way that can be integrated into regulatory/decision-making processes. Improved dialogue with policy-makers is needed to better understand research needs for policy-making and in the translation of EXPOsOMICS findings into understandable messaging.

Conclusion

The meeting concluded with feedback on the Exposomics project from the European Commission and the International Scientific Advisory Board. The potential of Exposome research to contribute to policy development includes: improved exposure assessment; enhanced specificity of actions to remove environmental hazards; identification of subgroups at risk; enhanced prediction and prevention of disease by early intervention; monitoring of policy results in reducing exposures; and elucidation of new hypotheses on the role of environment and health. Current limitations of this area were discussed in terms of showing added value for public health including the need to: improve communication of research results to non-scientific audiences and promote interaction among the producers and users of research; target research to the broader landscape of societal challenges – i.e. target priority policy areas that are under-researched or where current methods do not provide appropriate answers; and become better at drawing applicable conclusions – i.e. what is the added value of the research and what is the follow-up. Priorities for future work include the development and standardization of OMICS methodologies and technologies, data sharing and integration, and the demonstration of the added value of Exposome science over conventional approaches in answering priority policy questions. As a relatively new field, there will be a need to continue to demonstrate the utility of the Exposome approach to funders and policy makers.


References

Vineis, P, Chadeau-Hyam, M, Gmunder, H, Gulliver, J, Herceg, Z, Kleinjans, J. et al. 2016. The exposome in practice: design of the EXPOsOMICS project. Int J Hyg Environ Health. 220 (1 Pt A): 142-151.

Wild, CP. 2005. Complementing the genome with an “exposome”: the outstanding challenge of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol Biomarkers Rev. 14(8): 1847-50.

Rappaport, SM and Smith, MT. 2010. Environment and disease risks. Science. 330(6003):460-461.

Anon, 2016. http://exposomealliance.org

Vineis. P, Khan, AE, Vlaanderen, J. and Vermeulen, R. 2009. The impact of new research technologies on our understanding of environmental causes of disease: the concept of clinical vulnerability. Env Health 8:54.

Jerrett, M, Arain, A, Kanaroglou, P, Beckerman, B, Potoglou, D, Sahsuvaroglu, T, Morrison, J and Giovis, C. 2005. A review and evaluation of intraurban air pollution exposure models. Exp Anal Env Epidemio. 15, 185-204.

Hoek, G., Beele, R., de Hoogh, K., Vienneau, D., Gulliver, J., Fischer, P and Briggs, D. 2008. A review of land-use regression models to assess spatial variation of outdoor air pollution. Atm Env. 42(33): 7561-7578.

Van Roosbroeck, S., Wichmann, J, Janssen, NA, Hoek, G., van Wijnen, JH, Lebret, E and Brunekreef, B. 2005. Long-term personal exposure to traffic-related air pollution among school children, a validation study. Sci Total Environ. 268(2-2):565.73.

Klompmaker, JO, Montagne, DR, Meliefste, K, Hoek, G, and Brunekreef, B. 2015. Spatial variation of ultrafine particles and black carbon in two cities: results from a short-term measurement campaign. 508: 266-275.

Montagne, D, Hoek, G., Klompmaker, JO, Wang, M, Meliefste, K and Brunekreef, B. 2015. Land use regression models for ultrafine particles and black carbon based on short-term monitoring predict past spatial variation. Environ Sci Technol. 49 (14): 8712-8720.

Gulliver, J. and Briggs, DJ. 2005. Time-space modeling of journey-time exposure to traffic-related air pollution using GIS. Environ Res. 97(1): 10-25.

Chadeau-Hyam, M., Campanella, G., Jombart, T., Bottolo, L., Portengen, L., Vineis, P., Liquet, B. and Vermeulen, R.C.H. 2013. Environ Mol Mutagensis. 54(7):542-557.

Agier, L., Portengen, L., Chadeau-Hyam, M., Basagna, X., Giorgis-Allemand, L., Siroux, V., Robinson, O., Vlaanderen, J. et al. 2016. Environ Health Perspect. 124(12): 1848-1856.

Maron, DM and Ames, BN.1983. Revised methods for the Salmonella mutagenicity test. Mut Res/ Environ Mut and Related Subjects, 113(3-4): 173-215.

van Leeuwen, DM, van Herwijnen, MHM, Pedersen, M, Knudsen, LE, Kirsch-Volders, M., Sram, RJ, Staal, YCM, Bajak, E. and Kleinjans, JCS. 2006. Genome-wide differential gene expression in children exposed to air pollution in the Czech Republic.600 (1-2): 12-22.

Chadeau-Hyam, M., Athersuch, TJ, Keun, HC, De Iorio, M., Ebbels, TMD, Jenab, M, Sacerdots, C., Bruce, SJ, Holmes, E and Vineis, P. 2011. Meeting-in-the-middle using metaboliv profiling – a strategy for the identification of intermediate biomarkers in cohort studies. Biomarkers J. 16(1):83-88.

Lan, Q., Zhang, L., Li, G., Vermeulen, R., Weinberg, RS, Dosemeci, M and Rappaport, SM. 2004. Hemtotoxicity in workers exposed to low levels of benzene. Science, 306(5702): 1774-1776.

Smith, M., Vermeulen, R., Li, G., Zhang, L., Lan, Q., Hubbard, AE et al. 2005. Use of ‘omic’ technologies to study humans exposed to benzene. Chem Biol Interact, 153-154: 123-127.

Perera, F., Tang, W., Herbstman, J, Tang, D., Levin, L., Miller, R. and Ho, S. 2009. Relation of DNA Methylation of 5’-CpG Island of ACSL3 to transplacental exposure to airborne polycyclic aromatic hydrocarbons and childhood asthma. PLoS One. 4(2):e4488.

Zhang, Y., Li, C., Feng, R., Zhu, Y., Wu, K., Tan, X. and Ma, L. 2016a. The short-term effect of ambient temperature on mortality in Wuhan, China: a time-series study using a distributed Lag Non-linear model. Int J Environ Res. 13(7): 722.

Zhang, Y., Best, J., Tang, W., Tso, LS, Liu, F., Huang, S., Zheng, H., Yang, B., Wei, C. and Tucker, JD. 2016b. Transgender sexual health in China: a cross-sectional online survey in China. Sex Transm Infect. Pii: sextrans-2015-052350.

Bictash, M., Ebbels, TM, Chan, Q, Leng Loo, R., Yap, IKS, Brown, IJ, de Iorio, M., Daviglus, ML, Holmes, E. and Stamler, J. 2010. Opening up the “Black Box”: metabolic phenotyping and metabolome-wide association studies in epidemiology. J Clin Epidemiol. 63(9): 970-979.

Nicholson, G., Rantalainen, M., Maher, AD, Li, JV, Maolmodin, D., Ahmadi, KR., Faber, JH et al. 2011. Human metabolic profiles are stably controlled by genetic and environmental variation. Mol Syst Biol. 30(7):525.

Wang-Sattler, R., Yu, Y., Mittelstrass, K., Lattka, E., Altmaier, E., Gieger, C. et al. 2008. Metabolic profiling reveals distinct variations linked to nicotine consumption in human – first results from the KORA study. PLoS One. 3(12):e3863.

Holmes, E., Leng Loo, R., Stamler, J., Bictash, M., Yap, YKS, Chan, Q. et al. 2008. Human metabolic phenotype diversity and its association with diet and blood pressure. Nature. 453: 396-400.

Calderón-Garciduenas, L., Reed, W., Maronpot, RR, Henriquez-Roldan, C., Delgado-Chavez, R., Calderon-Garciduenas, A. et al. 2004. Brain inflammation and Alzheimer’s-like pathology in individuals exposed to severe air pollution. Toxicol Pathol. 32(6):650-8.

Wolf K, Stafoggia M, Cesaroni G, Andersen ZJ, Beelen R, Galassi C, et al. 2015. Long-term exposure to particulate matter constituents and the incidence of coronary events in 11 european cohorts. Epidemiology 26:565-574.

Uzoigwe JC, Prum T, Bresnahan E, Garelnabi M. 2013. The emerging role of outdoor and indoor air pollution in cardiovascular disease. N Am J Med Sci 5:445-453.

Newby DE, Mannucci PM, Tell GS, Baccarelli AA, Brook RD, Donaldson K, et al. 2015. Expert position paper on air pollution and cardiovascular disease. Eur Heart J 36:83-93b.

Liquet, B., Cao, KAL., Hocini, H. and Thiebault, R. 2012. A novel approach for biomarker selection and the integration of repeated measures experiments from two assays. 13:325.

Jacquemin, B., Kauffmann, F., Pin, I., Le Moual, N., Bousquet, J., Gormand, F., Just, J et al. 2009. Air pollution and asthma control in the Epidemiological study on the Genetics and Environment of Asthma. J Epidemiol Community Health. 66(9):796-802.

Kunzli, N., Bridevaux, PO, Liu, LJ, Garcia-Esteban, R., Schindler, C., Gerbasse, MW et al. 2009. Traffic-related air pollution correlates with adult-onset asthma among never-smokers. Thorax. 64(8):664-70.

Jacquemin, B., Siroux, V., Margaux, S., Carsin, A-E, Schikowski, T., Adam, M., et al. Environ 2015. Ambient air pollution and adult asthma incidence in six European Cohorts (ESCAPE). Env Health Perspectives. 123(6):613-621.

Huan T, Esko T, Peters MJ, Pilling LC, Schramm K, Schurmann C, et al. 2015. A meta-analysis of gene expression signatures of blood pressure and hypertension. PLoS Genet. 11:e1005035.

Hanson, MA and Gluckman, PD . 2014. Early developmental conditioning of later health and disease: Physiology or pathophysiology? Physiol Rev 94:1027-1076.



List of Websites:
Exposomics’ Project website address: www.exposomicsproject.eu/