Final Report Summary - HELIX (The Human Early-Life Exposome – novel tools for integrating early-life environmental exposures and child health across Europe)
Executive Summary:
The HELIX Project aimed to characterise the early-life exposome by assessing exposure to many environmental hazards during pregnancy and childhood, and linking these to children’s molecular omics signatures and to their cardiometabolic health, respiratory and immune health, and neurodevelopment (Vrijheid et al 2014). HELIX has successfully generated geospatial exposure estimates for outdoor air pollutants, noise, green space, meteorological and urban environment factors in around 30,000 mother-child pairs from 6 European birth cohorts during pregnancy and childhood. In 1301 cohort subjects HELIX also generated estimates for prenatal and postnatal exposure to chemical pollutants using highly sensitive biomonitoring methods. In total, over 200 separate exposure variables were measured. Complete molecular profile data sets comprising of urinary and serum metabolomics, plasma proteomics, blood cell DNA methylation, transcriptomics and miRNA data were generated for the same children. Harmonized child health outcome data were collected by applying common fieldwork protocols in the 6 countries. In doing this, HELIX has demonstrated that it is possible to build an early life exposome database with completely comparable biomonitoring data, geospatial data, child health outcome data, and omics signatures in the same subjects using a large-scale prospective framework.
First results focus on describing the Exposome, its correlations and its determinants, and highlight that: 1) exposure to most of the chemicals measured was abundant, with many being detected in over 90% of participants; 2) a considerable proportion of pregnant women and children are exposed to levels of urban exposures above international guidelines; 3) the exposome showed considerable variability across Europe for both urban and chemical exposures, confirming that location is a strong determinant of one’s personal exposome; 4) correlations within the same exposure family can be high, but correlations between exposures from different families were low, supporting the notion that epidemiological studies focusing on a single family of exposures may not be confounded by exposures from other families; 5) the exposome proved to be highly-dimensional and cannot be reduced to a small set of principal components; and 6) panel studies quantifying within-person variability in biomarker and outdoor exposures conclude that repeat biospecimen collections and personal dosimeters are important tools to refine exposome assessments and should form an integral part of future exposome studies.
An important challenge in associating the exposome with health is the simultaneous consideration of many correlated exposures. HELIX developed a statistical approach to evaluate Exposome-health associations in the light of complex correlation patterns, recognising that statistical techniques are often limited in their ability to efficiently differentiate true predictors from correlated covariates (Agier et al 2016, Barrera et al 2017). Within this framework, the systematic evaluation of child health risks related to multiple exposures will ultimately help guide public health efforts by allowing us to intervene on those chemical agents or urban and lifestyle exposures that are most likely to be associated with child health; indeed, preliminary results indicate a number of suspect exposures, including lead and reduced green space in relation to birth weight and maternal smoking in relation to child obesity. Also, HELIX is developing a catalogue of omics signatures associated with multiple environmental exposures, in order to better understand molecular mechanisms and early signs of damage, and potentially identify biomarkers; first results indicate that specific exposures tend to be associated with specific molecular signatures, e.g. methylation signatures for smoking, and metabolomics signatures for heavy metals and pesticides.
The database and scientific evidence created by HELIX will help to disentangle the relationship between multiple environmental exposures and health outcomes. To translate this scientific evidence into decision-making processes will require the development of tools like Health Impact Assessment (HIA). During the HELIX project, the HIA approach was used to estimate the child health impacts of seven individual environmental exposures in the EU, identifying air pollution as having the main impact.
By putting together over 200 exposures and omics and health data in a large, prospective early life framework, HELIX confirms that many environmental factors constitute a hazard for children in Europe. HELIX paves the way for a better characterization of the overall influence of environmental factors on human health.
Project Context and Objectives:
Background
The "exposome" concept encompasses the totality of non-genetic exposures from conception throughout the life course, complementing the genome. The exposome concept carries the expectation that the use of holistic and data-driven approaches, similar to those pioneered in the genomics fields, can result in advances in our understanding of the complex environmental component of disease aetiology. The exposome has been delineated to include three overlapping and complementary domains: 1) a general external domain including macro-level factors such as climate, urban environment and societal factors; 2) an individual external domain including agents such as environmental pollutants, tobacco smoke, diet and physical activity; and 3) a specific internal domain including gene expression, inflammation, and metabolism, often assessed through high-throughput molecular omics methodologies such as transcriptomics, proteomics and metabolomics.
The developing fetus, infant and child may be especially vulnerable to effects of environmental exposures since these are periods of rapidly growing and developing organs, immature metabolism, and the received dose relative to bodyweight may be greater than later in life. In utero and early life exposure to environmental stressors during critical windows can disrupt developmental processes and under The Developmental Origin of Health and Disease hypothesis such effects may permanently alter body structure, metabolism and physiology, and be expected to have a lifetime impact. Therefore the “early life exposome” is key, both as a starting point to develop a lifetime exposome and due the heightened impact of the exposome at this time. There is now moderate to good evidence for the effects of prenatal exposure to environmental contaminants, including air pollutants, polychlorinated biphenyls (PCBs), lead, mercury and organophosphate pesticides on fetal growth, neurological development, and the respiratory and immune systems. Evidence is also growing for effects on childhood growth, obesity, and metabolic signaling. At the same time it is clear that, up to now, the environment and child health field has almost uniquely focused on single exposure-health effect relationships. Only a few studies so far simultaneously considered more than a couple of families of compounds, focusing health outcomes such as birth weight, fecundity or type II diabetes mellitus.
Objectives
The HELIX project aims to measure and describe multiple environmental exposures from the three different exposome domains during early life (pregnancy and childhood) and associate these with omics markers and child health outcomes. The background and rationale of the HELIX project have been fully described (Vrijheid et al 2014). The objectives of HELIX were defined as follows:
Step 1: Measuring the external exposome
- To obtain estimates of exposure to persistent and non-persistent pollutants in food, consumer products, water and indoor air, during pregnancy and in childhood.
- To obtain estimates of chemical and physical exposures in the outdoor environment during pregnancy and in childhood: ambient air pollution, ambient noise, ultraviolet (UV) radiation, temperature, and built environment/green space.
Step 2: Integrating the external and internal exposome:
- To define multiple exposure patterns in the individual and outdoor environment, describe their predictors, and describe uncertainties and variability in the exposures assessed.
- To measure molecular signatures associated with multiple environmental exposures through analysis of profiles of metabolites, proteins, transcripts, and DNA methylation in biological samples from the children in the cohorts.
Step 3: Impact of the early-life exposome on child health
- To develop a novel multi-step statistical approach for the analysis of the association of patterns of multiple and combined exposures and child health outcomes.
- To provide exposure-response estimates for the association between multiple and combined exposures, and child health focusing on foetal and childhood growth and obesity, neurodevelopment, and respiratory health.
- To estimate the burden of common childhood diseases that may be attributed to multiple environmental exposures in Europe.
- To strengthen the knowledge base for European policy in the area of child and environmental health by engaging with, and effectively disseminating HELIX knowledge to, stakeholders including those responsible for risk management and mitigation and prevention strategies.
General Study Design
The HELIX study represents a collaborative project across six established and ongoing longitudinal population-based birth cohort studies in Europe: the Born in Bradford (BiB) study in the UK, the Étude des Déterminants pré et postnatals du développement et de la santé de l’Enfant (EDEN) study in France, the INfancia y Medio Ambiente (INMA) cohort in Spain, the Kaunus cohort (KANC) in Lithuania, the Norwegian Mother and Child Cohort Study (MoBa), and the RHEA Mother Child Cohort study in Crete, Greece. These cohorts were selected for participation in the HELIX project because: a) they could provide substantial existing longitudinal data from early pregnancy through childhood, b) they could follow-up children at similar ages, c) they could integrate questionnaires, biosampling and clinical examinations using common HELIX protocols, and d) they offered heterogeneity in terms of exposures and population characteristics.
Pregnant women in the original cohorts were recruited between 1999 and 2010. Three cohorts (INMA, KANC, RHEA) recruited during the 1st trimester of pregnancy, two through the 1st and 2nd trimesters (EDEN, MoBa), while in the BiB cohort women were recruited between weeks 26 and 28 of gestation (2nd/3rd trimesters). Inclusion and exclusion criteria varied between cohorts, as described in Table 1. All cohorts included at least one follow-up point during pregnancy, one at birth, and several after delivery.
Based on these six existing cohorts, HELIX used a multilevel study design, drawing on nested study populations for data collection of different intensities (Annex I, Figure 1): 1) the entire cohort in which factors arising primarily from outdoor exposures were assessed through geospatial models and linked to existing health outcome data; 2) a subcohort in which one new follow-up examination of the children between ages 6 and 11 years was carried out in order to assess child health outcomes and to fully characterize different areas of the exposome through questionnaires, biological sample collection, and biomarker and omics measurements; and 3) two panel studies in children and pregnant women to characterise in depth the variability in exposure biomarkers and omics biomarkers, individual exposure-related behaviours, and personal exposures.
The study population for the entire HELIX cohort includes 31,472 women who had singleton deliveries between 1999 and 2010, and for whom exposure to ambient air pollution during pregnancy had been estimated as part of the European Study of Cohorts for Air Pollution Effects (ESCAPE) project. The entire cohort includes nine regions from the six cohorts; we included only regions where geographic data were available to calculate air pollution levels and built environment indicators (Table 1). This meant, for example, that the city of Oslo and not the whole of the national MoBa cohort was included, and that only the Gipuzkoa, Sabadell and Valencia regions of the INMA study were included. In the other cohorts, women residing outside the main urban areas were excluded for the same reason.
In this study population, data on many variables had been collected in the individual cohorts during previous data collection points (during pregnancy and between birth and five years of age). Existing data included information on certain exposures (e.g.: maternal tobacco smoking during pregnancy, environmental tobacco smoke), key covariates (e.g.: pregnancy complications, maternal and child diet, maternal and child physical activity, child sleep, breastfeeding, other health related behaviours, indicators of socioeconomic status), and health and development outcomes. As part of HELIX, relevant datasets from all 31,472 mother-child pairs were transferred from the six cohorts to the central HELIX data warehouse located at the Barcelona Institute for Global Health (ISGlobal). Through data harmonisation these cohort-specific variables were converted to harmonised variables. This process involved summarising, checking, and matching the specific variable cohort-by-cohort and deciding a common coding system appropriate to each variable. Specific expert working groups throughout the HELIX consortium advised on the harmonisation rules for each variable. The child health and developmental outcomes harmonised as part of HELIX include birth outcomes, growth and obesity related outcomes, blood pressure, neurodevelopment, and respiratory health between birth and 5 years of age.
From the entire cohort, a subcohort of mother-child pairs was selected to be fully characterised for a broad suite of environmental exposures and “omics” data, to be clinically examined, and to have biological samples collected. A new follow-up visit was organised for these mother-child pairs between December 2013 and February 2016. Subcohort subjects were recruited from within the entire cohorts such that there were approximately 200 mother-child pairs from each of the 6 cohorts. Subcohort recruitment in the EDEN cohort was restricted to the Poitiers area and in the INMA cohort to the city of Sabadell.
Eligibility criteria for inclusion in the subcohort were: a) age 6-11 years at the time of the visit, with a preference for ages 7-9 years if possible; b) sufficient stored pregnancy blood and urine samples available for analysis of prenatal exposure biomarkers; c) complete address history available from first to last follow-up point; d) no serious health problems that may affect the performance of the clinical testing or impact the volunteer's safety (e.g. acute respiratory infection). In addition, the selection considered whether data on important covariates (diet, socio-economic factors) were available. Each cohort selected participants at random from the eligible pool in the entire cohort and invited them to participate in this subcohort until the required number of participants was reached. In total 1301 mother-child pairs with complete questionnaire and clinical examination data, and urine and blood samples, were included in the HELIX subcohort (Annex I, Figure 1). Several cohorts then invited and examined further subjects (N=322) following the same protocol, but these were not included in the measurement of exposure biomarkers for the HELIX study (Annex I, Figure 1).
The new follow-up visits for the subcohort took place in the six study centres at a local hospital, a primary care centre, or at the National Institute for Public Health (NIPH) in Oslo. During the follow-up examination, trained nurses interviewed the mothers, carried out health examinations of the children and collected biological samples using standardised operating procedures. Extensive cross-cultural questionnaires for use in all cohorts were developed and translated. These include questions required for accurate exposure assessment e.g. behavioral, diet and social data.
Intensive repeat panel studies collected data on short-term temporal variability in exposure biomarkers and omics biomarkers, individual behaviours (physical activity, mobility), and personal and indoor exposures. The child panel study included children from the HELIX subcohort (n=157, from all cohorts except MoBa) who lived in a first floor apartment or private house and were sampled following a maximum variation sampling strategy to high traffic-density exposure at home address. The pregnancy panel study included pregnant women from outside the cohorts in three cities, Barcelona, Grenoble, and Oslo (N=158). The inclusion criteria for these pregnant women were: to be 18 years or older at the start of pregnancy, to have a singleton pregnancy, to be living in the study area until delivery, and to have the first visit before the end of gestational week 20. Participants in the child panel study were followed for one week in two seasons, whereas in the pregnancy panel study the participants were followed for one week in two trimesters. In the child panel, the last day of the first week coincided with the subcohort examination, detailed above. Panel study subjects provided daily urine samples and at the end of each monitoring week blood samples were collected following the same procedures as for the subcohort. The panel study subjects wore smartphones for mobility and physical activity monitoring, electronic wristband UV dosimeters, active PM2.5 Cyclone samplers, and MicroAthelometers for continuous black carbon monitoring (Annex 1, Figure 2). .
Relevant datasets from all 31,472 mother-child pairs were transferred from the six cohorts to the central HELIX data warehouse located at ISGlobal. The data warehouse has been established in a format that allows future use beyond the project lifespan (2013-2017) as an accessible resource for collaborative research involving researchers external to the project. Access to HELIX data is based on approval by the HELIX Project Executive Committee and by the individual cohorts, who will evaluate potential overlap with ongoing work, adequacy of data protection plans, logistic and financial consequences, and adequacy of authorship and acknowledgement plans. We encourage interested researchers to contact us to set up collaborations. Further details on the content of the data warehouse and procedures for external access are described on the project website (http://www.projecthelix.eu/).
Project Results:
WP1 - Novel Tools for Individual Exposures
The aim of WP1 was to develop and apply novel tools and methods to obtain robust estimates of exposures to persistent and non-persistent pollutants in food, consumer products, water and indoor air, during pregnancy and early childhood. To improve the individual assessed exposures and reduce measurement error, sensitive methods for predicting exposure and repeated biomarker measurements, were applied. For all these exposures, methods that integrate exposure biomarkers and personal or environmental monitoring with knowledge on individual variability and behaviours, would lead to more accurate characterization of exposure. This required detailed characterization of short-term temporal variability and individual behaviours that influence exposure. The aim of WP1 was broken down in specific tasks;
A) To collect biological samples and exposure data in the HELIX subcohort;
B) To carry out the children and pregnancy exposome variability panel studies;
C) To determine biomarkers of exposure to persistent and non-persistent pollutants;
D) To develop and validate exposure prediction models for Disinfection By-Products (DBPs) and indoor air pollutants;
E) To assign exposure estimates to the subjects in the study and prepare a database with exposure
After years of thorough planning and field work the establishment of the HELIX subcohort as well as the panel studies were completed early in 2016. The final number of subjects who completed the fieldwork was 1623 children in the subcohort, and 157 children and 158 pregnant women in the panel studies respectively.
To reduce uncertainty and have as comparable results as possible, HELIX used one laboratory for all new measurements of the individual chemical exposure biomarkers. The samples were randomized into batches before chemical analyses. Data on biomarkers of exposure from pregnant women already existing were retrieved from the cohorts and undergone thorough quality control.
Biomarkers of contaminant exposure were measured in appropriate biological samples collected from the children at age 6-11 years and in samples previously collected from mothers during pregnancy or from the neonates during delivery (cord blood) and stored in cohort biobanks. Chemical assays were conducted in the laboratory at the Department of Environmental Exposure and Epidemiology at the NIPH, apart from analyses of metals/elements and cotinine, creatinine and blood lipids, which were subcontracted to ALS Laboratory Group Norway AS and Dr. Fürst Medisinsk Laboratorium AS, respectively. Biomarkers include: organochlorine compounds and brominated compounds, perfluoroalkyl substances and metals in blood, and non-persistent chemicals (phthalate metabolites, phenols, organophosphate pesticide metabolites, and cotinine) in urine samples (Table 2). Urine samples of the night before the visit and the first morning void on the day of the visit were combined to provide a slightly longer-term exposure assessment than can be achieved with one spot urine sample [M Casas et al - in review].
Table 3 gives and overview of the concentrations measured in samples taken from the mother during pregnancy and in the child 6-11 years later (the subcohort). This forms the basis for the individual early life chemical exposome to be included in work in the other work packages in HELIX. Figure 3 shows this “chemical exposome” for the persistent organic pollutants (the POPs exposome) measured in blood in children and mothers in the six cohorts. The concentrations of the exposure markers measured in the panel studies are presented in Annex I, Figure 4.
Concentrations of drinking water disinfection by-products (DBPs) during pregnancy were estimated from water company concentration and distribution data as part of the water contaminants and still birth, congenital anomalies, birth weight, preterm delivery (HiWate) project in four of the cohorts (BiB, KANC, INMA, RHEA). For EDEN and MoBa we followed the same methodology to obtain estimates during pregnancy. Data was not sufficiently complete to estimate child exposure to DBPs. Indoor air concentrations of nitrogen dioxide (NO2), particulate matter <2.5μm (PM2.5) particulate matter absorbance (PMabs), benzene, and toluene, ethylbenzene, xylene (TEX) were estimated by combining measurements in the homes of a subgroup of children during the two periods of the nested panel studies (see below) with questionnaire data from the subcohort.
Conclusions
This project has demonstrated that it is possible to perform a harmonized, though extensive, fieldwork using the same protocols for sample collection and clinical examination in six European mother-child cohorts. A wide range of exposure biomarkers were determined with high detection frequencies in all individuals in the entire subcohort. Thus, this study presents harmonized and completely comparable biomonitoring data for a plethora of environmental contaminants in children from several European countries, as well as comparable data for their mothers in samples taken during pregnancy. The concentrations found in the maternal samples were in general higher than concentrations measured in the children’s samples. For most of the persistent compounds the correlation between maternal and children’s levels was high, while considerably lower correlations were observed for most non-persistent compounds. The concentrations were significantly different between cohorts for more or less all compounds.
From the in depth, panel studies we have concluded that there is a substantial variability in exposure during pregnancy and over a year in school-age children for the majority of phthalate metabolites, phenols, and OP pesticide metabolites. The sample available for all HELIX subcohort children predicts well the annual exposure of most phthalates and oxybenzone, but it is not a good predictor of the other phenols and none of the OP pesticides. The quantification of the temporal variability of the non-persistent compounds can be used in Exposome studies to reduce bias. For most compounds 3-5 pools of 15/20 urines each would be necessary to obtain excellent reliability (Intraclass correlation coefficient>0.8).
Overall, HELIX has shown that children across Europe are exposed to a wide range of environmental chemicals in fetal life and childhood, and that the exposure varies between countries and by compounds. HELIX comprises a unique dataset to study exposure to chemical mixtures on an individual level and assess impact on health.
WP2 - Novel Tools for Outdoor Exposures
WP2 aimed to develop and apply novel tools and methods to obtain robust estimates of chemical and physical exposures in the outdoor environment ("the outdoor exposome"), focusing on key outdoor exposures (outdoor air pollutants, noise, green space, UV radiation).
In the entire cohort and subcohort, a GIS environment for the nine study areas was constructed, and, based on residential address histories, exposure estimates were assigned for ambient air pollutants, road traffic noise levels, surrounding (natural spaces green and blue spaces), built environment, ultraviolet (UV) radiation, and meteorological variables during pregnancy and childhood. These estimates build on existing land-use regression air pollution models (ESCAPE project), city noise maps, land use maps (“Urban Atlas” by European Environmental Protection Agency), raster maps of the normalized difference vegetation index (NDVI), raster maps of land surface temperature, building density, population density, connectivity, walkability and public bus transport map information for the built environment, and meteorological data, as described in more detail elsewhere [Robinson et al. – submitted]. Exposures where assigned within GIS techniques to all geocoded addresses of cohort subjects during the pregnancy, at birth, and at postnatal follow-up points. Data from existing regulatory monitors were used to back extrapolate ambient air pollution exposure models. The estimates for these outdoor exposures were calculated for the prenatal period and several postnatal periods up to the HELIX subcohort follow-up time point. Table 4 (Annex I) shows a summary of all GIS variables have been generated.
Daily measurements of temperature, humidity and pressure was obtained from a local weather station in each study area and averaged over pregnancy. Daily measurements of UV radiation (as erythemal UV, DNA damaging UV and vitamin D UV dose) at 0.5 x 0.5 degree resolution was obtained from the Global Ozone Monitoring Experiment onboard the ERS-2 (European Remote Sensing) satellite and averaged over pregnancy.
For assessment of air pollutants, including particulate matter with an aerodynamic diameter of less than 2.5 μm (PM2.5) and of less than 10 µm (PM10), nitrogen dioxide (NO2) and nitrogen oxides (NOX), we used land use regression (LUR) or dispersion models, temporally adjusted to measurements made in local back ground monitoring stations and averaged over the whole pregnancy period. For most cities we used site-specific LUR models developed in the context of the ESCAPE project. In Bradford, assessment for PM2.5 and PM10 was made based on the ESCAPE LUR model developed in the Thames Valley region of the UK and adjusted for background PM levels from monitoring stations in Bradford. The ESCAPE European -wide LUR model was applied for PM2.5 in Nancy, Poitiers, Gizpukoa and Valencia, corrected for local background monitoring data. In Gipuzkoa and Valencia, PM10 estimates were made based on local ratios to PM2.5 estimates. In Nancy and Poitiers, dispersion models were used to assess NO2 and PM10 exposure. In Rhea, new LUR models were developed for NO2 and NOX, incorporating new road traffic intensity variables collected from a new fieldwork campaign conducted at 80 monitoring points around the city (see supplementary information).
Noise levels Lden (day evening and night average) were derived from noise maps produced in each local municipality under the European Noise Directive. To improve comparability between centres, the values were categorized into six categories for analysis. In Heraklion, estimates on noise were newly modelled following new fieldwork (see supplementary information).
We followed the PHENOTYPE protocol to measure the surrounding vegetation, i.e trees, shrubs and parkland, and applied the Normalized Difference Vegetation Index (NDVI) derived from the Landsat 4–5 Thematic Mapper (TM) satellite images at 30m × 30m resolution (US Geology Survey, 2011). NDVI is an indicator of greenness based on land surface reflectance of visible (red) and near-infrared parts of spectrum and ranges between −1 and 1 with higher numbers indicating more greenness. To achieve maximum exposure contrast, we looked for available cloud-free Landsat TM images during the period between May and August for years relevant to our period of study and calculated greenness within 100, 300 and 500 meter buffers around each address. We calculated access to major green spaces (parks or countryside) and blue spaces (bodies of water) as the straight line distance from the home to nearest blue or green space with an area greater than 5000 m2 from topographical maps (urban atlas 2006 or local sources).
Topological maps for the following built environment indicators were obtained from local authorities or from Europe wide sources. Traffic density indicators (traffic density on nearest street, traffic load on major road within 100m and inverse distance to nearest major road) were calculated from road network maps following the ESCAPE protocol. Building density was calculated within 100 and 300m buffers by dividing the area of building cover (m2) by the area of buffer (km2). Population density was calculated as the number of inhabitants per km2 surrounding the home address. Street connectivity was calculated as the number of intersections inside 100m and 300m buffers, divided by the area (km2) of each buffer. Facility richness index was calculated as the number of different facility types present divided by the maximum potential number of facility types specified, in a buffer of 300 meters, giving a score of 0 to 1. Land use Shannon's Evenness Index (SEI) was calculated as the proportional abundance of each land use type multiplied by that proportion, divided by the logarithm of the number of land use types, in a buffer of 300 meters, giving a score of 0 to 1. We developed an indicator of walkability, adapted from the previous walkability indexes, calculated as the mean of the deciles of population density, street connectivity, facility richness index and land use SEI within 300m buffers, giving a walkability score ranging from 0 to 1.
Estimates were generated during pregnancy and follow up of the subjects in the main cohort, for all the children of the subcohort and for the panel studies of the children and pregnant women.
Results from estimates made during pregnancy showed that mean noise levels ranged from 49.6 dB (Kaunas) to 63.9 dB (Heraklion), mean NO2 levels ranged from 13.6 ug/m3 (Heraklion) to 43.2 ug/m3 (Sabadell), mean walkability score ranged from 0.22 (Kaunas) to 0.32 (Valencia) and mean Normalized Difference Vegetation Index (NDVI, 300m buffer) ranged from 0.21 in Heraklion to 0.51 in Oslo. NDVI was correlated with NO2 (-0.42) noise (-0.26) building density (-0.74) and UV radiation (-0.13) among other indicators. Four PCs explained more than half of variation in the urban exposome. We observed considerable heterogeneity in social patterning of the urban exposome across cities. For example, high SES women lived in greener, less noisy and polluted areas in Bradford (0.18 standard deviations of PC1, 95% CI: 0.15 0.22) but the reverse was observed in Oslo (−0.25 standard deviations of PC1, 95% CI: −0.32 −0.18).
Furthermore, participants of panel studies were monitored twice regarding their mobility, physical activity, and personal exposure to air pollution, noise, ultraviolet (UV) light, and natural outdoor environment, using a newly developed personal exposure monitoring (PEM) kit. The two repeated measurements were obtained during two separate normal weeks (i.e. school and working weeks) with 6 months of difference. The PEM kit was composed of: (i) a belt with an attached smartphone and accelerometer; (ii) a wrist UV dosimeter; (iii) a small backpack fitted with one gravimetric sampler to measure particulate matter with aerodynamic diameter of 2.5 µm or less (PM2.5); and one real-time sampler to measure black carbon (BC); both with the inlets attached to one handle of the backpack at the breathing zone; and (iv) two extra gravimetric samplers for PM2.5: one placed at the living room and the other outside of the main window or balcony of the house.
Geographical location was obtained from the ExpoApp application, which was running in a smartphone GT-S5360. Physical activity was assessed with a wGT3X-BT tri-axial accelerometer (ActiGraph, LLC, USA) and the ExpoApp. The wGT3X-BT and ExpoApp were set to sample at 30 and 10 Hz, respectively. The physical activity information obtained was the wearing time, intensity, duration, and frequency of the physical activity at minute resolution (Choi et al. 2012; Crouter et al. 2010, 2013; Donaire-Gonzalez et al. 2013). The intensity of physical activity together with the age, sex and weight of individuals was used to estimate the inhaled rate per minute, using the existing equations from the Environmental Protection Agency
We found that there is considerable variation in the personal urban early life exposome with a considerable percentage exposed to high levels. More than 10% of pregnant women and children were exposed to fine particles levels ≥ 25µg/m3 and ≥ 50% of them did not have any contact with NOE during one week. Moreover, children exposure levels overall were higher than those in pregnant women, particularly for UV-B doses and physical inactivity. Furthermore, we found that most of the personal exposures were not correlated but the relationships were different in children than in pregnant women. Finally most of the variability of the exposures was explained by city characteristics, especially in children.
Conclusions
We estimated and measured successfully a large range of exposures including air pollution, UV, green space, built environment, meteorology, and noise for the participants at different time points. The results show that there is considerable variation in the exposure levels between subjects and cities depending on the exposure of interest. The estimates will be used in the epidemiological analyses.
WP3 - Integrating Multiple Exposures
WP3 aimed to integrate data on multiple exposures, exposure determinants, and exposure variability (temporal, individual, toxicokinetic), in order to define multiple exposure patterns and describe exposure uncertainties.
Predicting target tissue dose through PBPK modelling
The objective of this task was to analyze the measured biomarker concentrations (from HELIX subcohort and panel studies) using physiologically based pharmacokinetic (PBPK) models to simulate the dosimetry in the target tissues. Our methodology was applied to compounds having different half-lives: PFOS and PFOA for semi-persistent compounds applied to the subcohort of 1,200 mother-child pairs, and a phthalate (DEHP, di(2-ethylhexyl) phthalate) for non-persistent compounds applied to the panel study of pregnant women (about 150 women).
PFOS: We chose to adapt a generic and lifetime PBPK model to describe the toxicokinetics of PFOS (Beaudouin, et al. 2010). Several changes were made to fit the model to the characteristics of PFAS and to improve the physiological description. The exposures of the mother and child were reconstructed using the measured biomarkers and the PBPK model. A high diversity of toxicokinetic profiles has been estimated based on the individual information collected through the questionnaires. The main determinants of the child exposure are: the levels at birth (correlated with the mother’s biomarker), the breastfeeding that can result in high levels in child plasma, and the measured biomarker itself. Using the reconstructed exposure scenarios for the children as input, PBPK models were used to simulate target tissue dosimetry at critical time periods. Our results show that the in utero indicators are all highly correlated with the measured maternal plasma concentration at the time of pregnancy, and that actual measurements during pregnancy and at the age of 6 to 9 do not correlate well with the predicted internal exposure during the first years of life (birth to 4 years).
DEHP: A toxicokinetic model (Lorber et al., 2010) that simulates serum and urine concentrations of DEHP and five metabolites (MEHP, MECPP, MEHHP, MEOHP, 2cx-MMHP) using a first order dissipation equation along with routing assumptions from parents to metabolites and to the bladder for excretion via urination was used. This model was used in a reverse dosimetry approach for estimating the distribution of exposure levels in the environment that could give rise to measured biomarker concentrations in a population. In our study, estimation of DEHP exposure by reverse dosimetry was performed for each pregnant woman. We estimated by reverse dosimetry the exposure to DEHP (in µg/kg/d) that may lead to the concentrations of DEHP metabolites measured in our study for three different scenarios of exposure. We obtained distributions of estimated exposure for each of the 3 following scenarios. The reverse modelling showed that, the mean daily intakes (DI) estimated for the women from Barcelona and from Oslo are quite similar for all the scenarios. The diet exposure scenario shows a higher variability than the two other scenarios. The diet exposure scenario seems to better depicts the high variability of the DEHP metabolites urinary concentrations. The MEHP blood concentration predicted for the continuous exposure is constant over time whereas the concentration really variable for the two other scenarios (day exposure and diet exposure). The maximal concentration (Cmax) of MEHP serum concentration predicted is higher for the diet exposure than for the continuous. There is a factor of 4 between Cmax predicted for diet and for continuous scenarios. This major difference between scenario predictions is not observed for the Area Under the curve (AUC) of MEHP serum concentration. In fact, the values of the AUC estimated are relatively similar whatever the scenario considered. The diet exposure seems to better depict the variability of the data. To conclude, toxicokinetic model and questionnaire with diet exposure data and urination times are crucial elements for the DEHP exposure assessment and for internal DEHP metabolite prediction.
Analysis of variability and uncertainties in exposure assessments
Non-persistent exposures: To estimate the uncertainty in assessments of exposure to non-persistent pollutants and second-tobacco smoke exposure we used data from the children and pregnancy panel studies. The specific objectives were: a) to evaluate between-trimester variability in pregnant women and within-week and between-season variability in school-age children of urinary concentrations of phthalate metabolites, phenols, organophosphate (OP) pesticide metabolites, and cotinine; b) to estimate whether the last pool of 2 urines of the first follow-up week (measure available in the subcohort children) was a good measure of the mean annual concentration in children; c) to calculate the number of biospecimens needed to obtain excellent reliability (defined as an ICC of 0.80 or more) of each chemical.
In children, we determined non-persistent pollutants in the urine for 4 consecutive days’ pools of week 1 (night + first morning), in the 2 weekly pools (of all samples), and in 1 pool of night and first morning voids of the last day of week 2. In pregnant women we determined non-persistent pollutants were determined in the 2 weekly pools, and phthalates were also determined in all the first morning and night voids of the first week in 45 of these 157 women (15 from each city).
Using these samples we observed that:
a) The HELIX subcohort measurement, which corresponds to the pool of night and first morning voids of the last day of week 1, is a good predictor of exposure in the week before of the majority of phthalates and phenols, DMTP and DEP pesticides, and cotinine. Subcohort measurement however, is not a good predictor of exposure in the week before of BUPA and of the majority of OPs. The subcohort measurement (week 1) is a good predictor of exposure in the second week (approximately 6 months of difference) of the majority of phthalates, BUPA, benzophenone-3 (OXBE), and cotinine but it is not a very good predictor of the majority of phenols, particularly BPA, and OPs.
b) Phthalates (except oxoMiNP), benzophenone-3, and cotinine have a low seasonal variability in children whereas the other phenols and OPs concentrations vary a lot between seasons. In pregnant women we generally observed a high variation of non-persistent pollutants concentrations between the 2nd and 3rd trimesters of pregnancy.
c) Phthalates (except oxoMiNP) and phenols (except BUPA) have low between-day variability whereas OPs have a high between-day variability in children. In pregnany women MEP, MiBP, MnBP, and MBzP have low between-day variability whereas all the other phthalates have high between-day variability.
For most non-persistent chemicals, three daily pools of 2 urines each would be needed to obtain excellent reliability for a weekly exposure window. Four weekly pools of 15-20 urines each would be necessary to obtain excellent reliability for a yearly or pregnancy exposure windows.
The ICCs calculated for each pollutant can be integrated in the exposure-response models by using the regression-calibration method. This regression allows correcting for bias from measurement error in generalized linear models.
Outdoor exposures: To estimate the uncertainty of outdoor exposures we also used data from the children and pregnancy panel studies. Specific objective was to characterize the intra- and inter-participant and intra- and inter-city variability on the personal exposure to air pollution, noise, ultraviolet (UV) light, and built/natural environments in childhood and during pregnancy. From five out of six cohorts, 150 children aged 6-9 were sampled following a maximum variation sampling strategy to high traffic-density exposure at home address. From three of the cohort regions, 150 volunteer women were recruited during her first trimester of pregnancy. With 6 months of difference, participants were continuously monitored for a normal week regarding its mobility, physical activity, and personal exposure to air pollution, noise, ultraviolet light, and built/natural environment, using a personal exposure monitoring (PEM) kit. Our findings reflect that:
a. Exposure to air pollution of children and pregnant women is different because children are more exposed to particulate matter, while pregnant women are more exposed to black carbon.
b. Exposure to air pollution is determined mainly by the activity patterns of participants’ daily life rather than by individual or city characteristics.
c. The need of epidemiological studies to include the pattern of daily life activity of participants into their air pollution exposure to mitigate the exposure misclassification.
d. The need for research to include personal exposure assessment and the importance of having access to accurate, inexpensive and not burdensome personal exposure tools.
Factor analysis of multiple exposure patterns
The objective of this task was to define exposure patterns in the multiple exposure data collected in WP 1 and WP2, using a factor analysis approach that will create a reduced set of continuous exposure indices, each of them composed of exposures that tend to occur simultaneously in the population.
The outdoor exposome collected in the HELIX project was composed of 26 different exposures belonging to 7 families of exposures (noise, air pollution, traffic, meteorological variables, built environment, green space and blue space), and was available for 27,921 pregnant women form 9 cities (Bradford, United Kingdom; Poitiers and Nancy, France; Sabadell, Valencia and Gizpukoa, Spain; Kaunas, Lithuania; Oslo, Norway; Heraklion, Greece). The full exposome was composed of 208 exposures belonging to 15 families of exposures, and was available for 1285 participants. Exposure variables were transformed when needed to have a symmetric distribution. To maximize the use of available information and provide unbiased estimates, we conducted multiple imputation of missing values using the method of chained equations. First, we displayed correlation heat maps for all exposures, overall and by cohort. Subsequently, we standardized exposures by dividing them by their respective standard deviation. Those transformed variables were used to conduct a principal component analysis. We calculated the percent of explained variance as a function of the number of principal components and displayed it graphically. The number of principal components to be retained was selected upon exploration of the latter graphical display. Varimax rotation was applied to the final solution. The loadings of each principal component were examined in order to label each principal component. The results of these analyses highlighted the following points:
a) Although correlations within the same exposure family can be high, correlations between exposures from different families were low. This supports that epidemiological studies focusing on a reduced set of exposures may not be confounded by having omitted exposures from other families.
b) The exposome is high-dimensional, as it cannot be reduced to a small set of principal components. Even the outdoor exposome required 15 components to explain most of the variability. In the full exposome, up to 60 components were needed to explain most of the variation.
c) Between-city differences influence the correlations. Correlations between exposures tend to be larger when city is not controlled for, as they also reflect between-city correlations. The first principal components reflect the differences between cities.
Model-based clustering approach to identify groups of subjects with similar pattern of exposure
The aim this task was to identify clusters of participants sharing a similar exposome. Continuous variables were transformed to achieve symmetric distributions. Since we were not interested in capturing between-cohort correlations (e.g. southern cohorts having warmer temperatures and higher air pollution), we subtracted cohort means from each exposure prior to the analyses. Continuous variables were standardized by dividing by the global standard deviation. The dataset contained continuous and binary variables, so clustering techniques that allow for mixed data were used. Several clustering techniques were applied to the data, as there is no single method that is guaranteed to outperform the other methods in all instances. In particular, we used Bayesian model based clustering, as implemented in the R statistical package PReMiuM1; partitioning around mediods (PAM) based on Gower distances, which accommodate continuous and binary variables (https://www.r-bloggers.com/clustering-mixed-data-types-in-r/); and hierarchical clustering based on Gower distance, as implemented in the R function hclust(). PReMiuM finds the optimal number of clusters using the Silhouette width, and it also performs variable selection. For the PAM method, we also calculated the Silhouette width. For the hierarchical clustering, we used the bootstrap method implemented in the pvclust R package.
In summary, the three clustering methods used did not provide useful solutions for the ultimate aim of the HELIX project, which was to link exposure clusters to health outcomes. Two-cluster solutions do not provide enough variability to capture different exposure patterns. Besides, they were highly collinear with cohort, and our analyses on the effects on health will adjust for cohort. The solutions of model based clustering, which provided a large number of clusters, were not useful either, as the clusters had too small sample sizes, and the solutions were again highly collinear with cohort. The conclusion from this part of the analysis is that clustering of exposures may not be adequate for the following analyses on the effects of the exposome on health outcomes. Other techniques, such as the principal component analysis conducted in the previous task may provide better alternatives.
Conclusions
WP3 focused on the integration of data on multiple exposures, exposure determinants, and exposure variability (temporal, individual, toxicokinetic), in order to define multiple exposure patterns and describe exposure uncertainties. Main conclusions of this WP are:
1) The analysis of the biomarkers and questionnaire data with PBPK models allows reconstructing the external exposure of semi-persistent compounds (PFAS) in children and simulating target tissue dosimetry at critical time periods during early life. Our results showed that an adequate use of individual information (e.g. breastfeeding) with a PBPK model can rebuild realistic exposure scenarios. Accounting for this inter-individual variability can reduce uncertainty and may increase the power of association analyses between adverse effects and exposure. For non-persistent compounds (DEHP), we highlighted the importance of individualizing the exposure scenario to adequately estimate the maximum exposure of the individuals at certain times of the day.
2) In terms of the characterisation of uncertainties and variabilities, for most non-persistent chemicals, multiple pools of multiple urines would be needed to obtain excellent reliability in exposure assessment; for example, 4 pools of 15-20 urines each would be needed to get a reliable estimate over the entire pregnancy. The variability in environmental outdoor exposures was mainly determined by the activity patterns of participants’ daily life rather than by individual or city characteristics. These results confirm the need of using personal exposure assessment methods for outdoor exposures in epidemiological studies and the importance of having access to accurate, inexpensive and not burdensome personal exposure tools.
3) In terms of exposure patterns: The exposome is high-dimensional, as it cannot be reduced to a small set of principal components or clusters; In fact, clustering of exposures may not be adequate for multicentre analyses on the effects of the exposome on health outcomes as location is a strong determinant of one’s personal exposome.
4) Although correlations within the same exposure family can be high, correlations between exposures from different families were low. This supports that epidemiological studies focusing on a reduced set of exposures may not be confounded by having omitted exposures from other families.
Future Exposome studies should continue refining exposure assessment through repeated collection of biospecimens and personal dosimeters in even larger populations.
WP4 - Integrating Molecular Exposure Signatures
The objective of WP4 was to determine molecular signatures associated with environmental exposures through analysis of profiles of metabolites, proteins, RNA transcripts, and DNA methylation. Specific objectives were:
▪ To generate molecular profiles in biological samples from the cohort studies using optimised protocols
▪ To correlate specific exposures and exposure clusters to molecular profile data
▪ To integrate information from molecular profiles using pathway analysis approaches to define biological pathways associated with exposure
Omics data generation and quality control
Urinary metabolic profiles were analysed on a 14.1 Tesla (600MHz 1H) nuclear magnetic resonance spectrometer at Imperial College London (ICL). Most samples analysed (1273/1366) were a pool of samples taken from each child in the morning and the night before. The targeted metabolomics AbsoluteIDQTM p180 Kit (BIOCRATES Life Sciences AG) was used to profile the serum samples from the sub-cohort (n=1364 samples). The kit allows the targeted analysis of 188 metabolites in the metabolite classes of amino acids, biogenic amines, acylcarnitines, glycerophospholipids, sphingolipids and sum of hexoses.
Buffy coat DNA was extracted using the Chemagen kit (Perkin Elmer) in batches of 12 samples. Samples were extracted by cohort. DNA concentration was determined in a NanoDrop 1000 UV-Vis Spectrophotometer (ThermoScientific) and with Quant-iT™ PicoGreen® dsDNA Assay Kit (Life Technologies). 700 ng of DNA were bisulfite-converted using the EZ 96-DNA methylation kit following the manufacturer’s standard protocol, and DNA methylation was assessed with the Infinium HumanMethylation450 beadchip from Illumina, following manufacturer’s protocol. DNA methylation data was pre-processed using the minfi package . After sample and probe quality control ), data was normalized with the functional normalization method, which also includes Noob background subtraction and dye-bias correction . The final dataset consisted of 1,347 HELIX samples representing 1,192 subjects and 485,512 probes (or 386,518 probes after filtering of probes with SNPs, probes that cross-hybridize and probes in sexual chromosomes).
RNA was extracted from 1,382 HELIX samples (and 308 extra HELIX samples) using the MagMAX for Stabilized Blood Tubes RNA Isolation Kit (ThermoFisher). Mean values for the RIN, concentration (ng/ul) and Nanodrop 260/230 ratio were: 7.05 109.07 and 2.15. Gene expression, including coding and non-coding transcripts, was assessed with the Affymetrix Human Transcriptome Array 2.0 ST arrays (HTA 2.0). Amplified and biotynylated sense-strand DNA targets were generated from total RNA of the 1,304 samples with good RNA quality and hybridized on Affymetrix HTA 2.0 arrays. Data was normalized with the GCCN (SST-RMA) algorithm at the gene and transcript level. Annotation to transcripts clusters was done with the ExpressionConsole software using the HTA-2_0 Transcript Cluster Annotations Release na36. Control probes and probes in sexual chromosomes or probes without chromosome information were excluded. Probes with a call rate <70%, based on DABG (Detected Above Background) p value < 0.05 were excluded from the analysis.
Expression miRNA levels of 1,087 samples with good RNA quality were analysed using the SurePrint Human miRNA Microarray rel. 21 (Agilent) following Agilent's recommendations. Briefly, RNA samples were concentrated or evaporated in order to reach the required concentration using a vacuum equipment (SpeedVac). The miRNA Complete Labeling and Hyb kit generates fluorescently-labeled miRNA with a sample input of 100 ng of total RNA. This method involves the ligation of one Cyanine 3-pCp molecule to the 3' end of a RNA molecule. Agilent SurePrint G3 Human miRNA microarrays were hybridized following the Agilent Microarray Hybridization Chamber User Guide. After an initial quality control based on laboratory parameters, miRNA levels were normalized using the least variant set (LVS) method with background correction using the Normexp method in limma package . After normalization, miRNAs with a call rate <70%, based on Agilent’s p value, and miRNAs in sexual chromosomes were filtered out. The final dataset consisted of 1,078 samples and 330 miRNAs.
A set of 43 proteins were selected a priori based on the literature and on the Luminex kits commercially available from Life Technologies and Millipore. Three kits were selected, which assessed a total 50 measurements: Cytokines 30-plex (Cat #. LHC6003M), Apoliprotein 5-plex (LHP0001M) and Adipokine 15-plex (LHC0017M). For protein quantification, an 8-point calibration curve per plate was performed with protein standards provided in the Luminex kit and following the procedures described in the standard procedures described by the vendor. The % of coefficients of variation (% CV) for each protein estimated by plate and then averaged ranged from 3.42% to 36%. Seven proteins were removed because they had <30% of measurements in the linear range. For the 36 proteins that passed the QC, data was log transformed to reach normal distribution. Then, the plate batch effect was corrected by subtracting for each individual and each protein the difference between the overall protein average minus the plate specific protein average. Finally, values >LOQ2 (upper limit of quantification) were set to NA and values
For further details including procedures for data quality control please see the previous deliverable report (see Deliverable Report D4.4 “Completed omics database, selection of targets for health analysis”).
ExWAS analyses
Linear regression models were generated for each exposure-omics biomarker pair using omicRexposome, which was developed as part of HELIX project (https://github.com/isglobal-brge/omicRexposome). Models were adjusted for age, sex and cohort. Surrogate variable analysis (SVA) was applied to correct for main unwanted variability derived from technical batch, blood cell type proportions or others. This approach was applied only on the following datasets: miRNA, gene expression and methylation. The significance threshold was adjusted for multiple testing using Bonferroni correction considering either the number of omics markers for each platform (‘O’) or the number of omics-exposure biomarker pairs (‘OE’).
Omics data: number of samples and features
The final number of samples and features for the HELIX subcohort –omics can be seen in Table 5,1 (Annex I). The overlap between samples with several omics data can be seen in Table 5.2 (Annex I); in total, 874 samples have available all omics data planned in HELIX in addition to the exposome and health data.
ExWAS: metabolomics, proteomics and miRNAs
Our initial ExWAS analyses focused on integration of the exposome with proteomics, metabolomics and miRNA profiles as these were available in advance of other omics and with less features require less computational expense to complete. ExWAS analysis revealed a number of significant associations between molecular features and exposures (Figure 5, Annex I), even after adjustment for multiple testing, considering different thresholds. The highest number of significant associations was found with the postnatal exposome, especially with the serum metabolome. Few associations (N=4) overlapped between pregnancy and postnatal exposome.
Figures 6-9 (Annex I) illustrate examples of the emergent exposure/biomarker signatures revealed by ExWAS analysis. Serum metabolites exhibited strong positive correlations to levels of PFASs as well as levels of arsenic (As) and mercury (Hg) (Figure 6). Examination of the associated metabolites indicated an over-representation of phosphatidylcholine lipids containing polyunsaturated fatty acids (nomenclature PC total number of carbons in fatty acids: total number of double bonds in fatty acids). Since humans cannot desaturate fatty acids beyond carbon 9, a key supply of these essential fatty acids is the diet. Polyunsaturated fatty acids levels are high in certain plant products and also fish and other seafood. The latter can accumulate toxins such as heavy metals and therefore the observed signature may be driven in part by common routes of exposure. A similar pattern was observed for urinary metabolome (Figure 7). Arsenic and mercury were associated with trimethylamine-N-oxide, a metabolite found in fish; while organophospate pesticides were associated with hippurate and proline betaine, metabolites incorporated from fruits and vegetables.
We also observed a cluster of inverse correlations between exposure to organochlorines (OCs), and to a lesser degree PBDEs, with circulating levels of the appetite-regulating hormone leptin, and two important pro-inflammatory, adiposity-associated cytokines (adipokines), IL-6 & IL-1beta (Figure 8). Associations were attenuated after adjustment for child body mass index, which suggests that fat mass acts as a confounder. Blood concentration of lipophillic persistent organic pollutants such as OCs and PBDEs is dependent on total body fat (ie. given the same exposure, obese children have lower blood concentration of OCs), and on the other hand fat mass is the main producer of adipokines.
Blood miRNAs were associated with meteorological exposures such as temperature and ultraviolet radiation (Figure 9). Further investigation is needed to discard potential cohort effects.
NO2, mercury and HCB: methylome and transcriptome
We are working on the analysis of the full exposome (ExWAS) and the methylome and transcriptome. At this point, we have finalized the analysis of three “reference” exposures: NO2, mercury and HCB. Postnatal NO2 levels were associated with blood DNA methylation levels at 44 CpG sites. These, were enriched in genes related to blood pressure and body mass index regulation. In contrast prenatal NO2 levels did not give many associations at age 7-9 years (n=4), and they did not overlap with associations observed for postnatal NO2. Exposure to mercury was not associated with methylation, while prenatal and postnatal HCB was associated with very few CpG sites (n=4-5). We did not find any statistically significant association between these three exposures and the transcriptome.
SUMMARY
In the HELIX project we have successfully generated complete molecular profile data sets comprising of urinary and serum metabolomics, plasma proteomics, blood cell DNA methylation, transcriptomics and miRNA data for 874 children in the HELIX subcohort, with up to 1198 children’s samples analysed for individual omics platforms. An EXposure Wide Association Study (ExWAS) was conducted with metabolomics, proteomics and miRNA data and the child-matched full exposome dataset from HELIX. These analyses identified several significant clusters of associations, including: serum polyunsaturated glycerophospholipids with exposure to heavy metals and PFASs; urinary trimethylamine-N-oxide with exposure to heavy metals and urinary hippurate and proline betaine with organophospate pesticides. These signatures are likely to represent common routes of exposure, e.g. fish and seafood as a common source of polyunsaturated fatty acids and metals. We also found associations between OCs and PBDEs with plasma adipokines, which can reflect a link with fat mass, e.g. fat mass as a driver of adipokine expression and storage of lipophilic chemicals. A similar statistical analysis followed with functional enrichment is ongoing for the methylome and transcriptome and the full exposome.
WP5 - Linking the Exposome to Child Health
WP5 aimed to characterize the effect of the Exposome on specific highly prevalent child health outcomes. These health outcomes are pre- and postnatal growth and obesity, asthma and respiratory function, and neurodevelopment. Objectives of WP5 were to carry out the health outcome examinations in the HELIX subcohort; to design and maintain the HELIX data warehouse; to perform a simulation study to compare the efficiency of various statistical approaches considered to assess the impact of the Exposome on human health; and to provide exposure-response estimates for the association between the Exposome and child health.
Health outcome examinations were carried out using standarised protocols in the six cohorts (children aged 6-11 years). Results show large differences between cohorts in the health outcomes: For example, food allergy questionnaires showed that overall 21% of children were reported to have at least one food allergy (ever experienced), ranging from 15.6% in RHEA to 35% in INMA. The percentage of children who had ever had asthma was low in INMA (3.6%) and high in BiB (18.5%) and EDEN (20.2%). Overall, 18.8% of children were overweight and 9.9% were obese (total 27.7%). The percentage of overweight and obese children (using the age-and-sex-standardised z-scores) was highest in RHEA (37.2%) and INMA (42.3%) and lowest in MoBa (15.8%). ADHD symptoms assessed through the Conner’s rating scale were classified using the cut-off score of the 80th percentile. Using this classification, 10.1% of children in the subcohort were classified as having ADHD symptoms, ranging from 4.4% in MoBa to 15.2% in KANC. The total problems score of the CBCL, which consists of the sum of ratings on all 120 behavioural and emotional items of the CBCL, also showed that mothers in MoBa reported the lowest total score (median score 9) and mothers in KANC the highest (median score 27).
The HELIX data warehouse was constructed as a relational database created in MySQL. New data, collected through the common protocols during the subcohort and panel study fieldwork, were entered directly into an electronic database and then uploaded into the data warehouse. Questionnaires were computer-based with a direct entry to the database. All data were locally and centrally checked by examination of the ranges, distributions, means, standard deviations, outliers and logical checks. Data outliers and missing values were checked with the local cohort field workers and, where possible and relevant, replaced by correct values. All new measurements of exposure biomarkers and omics from the labs, and all exposure variables estimated through geospatial models and other methods, were added to the data warehouse as they became available.
Exposome-health associations
From a methodological perspective, most previous studies relating the Exposome to health relied on the Environment-Wide Association Study (EWAS) 5, possibly followed by a multiple regression step 6. Several other regression-based methods exist and allow accounting for a potential joint action of multiple exposures on health. Sparse Partial Least Square (sPLS) for instance has recently been used in a study of male fecundity 4, while Elastic Net (ENET) was used to link multiple environmental contaminants to birth weight1. The statistical performances of these various models in an Exposome context remain to be systematically assessed. Two- or three-way interactions between environmental exposures have been described in the literature
and statistical methods to uncover interactions among a large set of exposures have been suggested 7,8. Again, their efficiency and limitations have not been systematically assessed.
Our aims were twofold: 1) to identify agnostic statistical methods most suited for the study of the effects of the Exposome on health, both in the absence or presence of interactions between the components of the Exposome; 2) to characterize the association between the early-life Exposome and child health (birth weight, body mass index, BMI, in childhood, respiratory health, neurodevelopment, blood pressure).
Methods
Our approaches combined simulation studies based on a realistic Exposome data structure and real analyses of the unique HELIX subcohort population, in which over 100 exposure factors assessed from biomarkers (WP1) and environmental models (WP2) and belonging to over 15 exposure families have been characterized during pregnancy and in childhood.
Real analyses were performed using the statistical models that showed the best performances in simulation studies, as well as univariate descriptive approaches (EWAS). All models were adjusted for a predefined set of potential confounding factors. All continuous variables were normalized and standardized by the IQR before being analysed. Effort was made to reduce the initial set of exposures and avoid correlation >90%; when exposures were assessed at different windows (e.g. time points or buffers), we a priori selected one of them in the primary analysis. Independent analyses were performed for the exposure variables measured during prenatal and postnatal periods. All analyses were performed using the R software (www.r-project.org). We used the Rexposome package for drawing plots, mice for multiple imputation and DSA for the DSA algorithm.
5.1 Simulation studies: which approaches should be used to characterize Exposome-health relations?
Regarding the simulation studies , in the absence of interaction, the elastic net, sparse partial least-squares regression, Graphical Unit Evolutionary Stochastic Search (GUESS) Bayesian variable selection model, and the deletion/substitution/addition (DSA) algorithm showed on average over all simulation settings a sensitivity of 78% and a FDP of 37%, with minor differences between methods. The environment-wide association study (EWAS) underperformed these methods in terms of FDP (68% on average) and bias, with a higher sensitivity. When we assumed that some exposure factors acted in synergy, the models GLINTERNET and DSA had better overall performance than the other methods, with GLINTERNET having better properties in terms of selecting the true predictors (sensitivity) and of predictive ability, while DSA had a lower number of false positives. In terms of ability to capture interaction terms, GLINTERNET and DSA had again the best performances, with the same trade-off between sensitivity and false discovery proportion 9.
5.2 Which components of the Exposome were related to the health of HELIX children?
Regarding the analyses of HELIX subcohort data, EWAS and DSA models identified several associations with the health outcomes considered, several of which were consistent with the existing literature.
Birth weight: In the case of the birth weight analysis, lead was the only exposure associated with birth weight in the DSA model corrected for exposure misclassification (an effect restricted to boys, corresponding to a decrease in mean birth weight with lead exposure, Figure 14, Annex I). The only interaction term identified was between a marker of socio-economic status and blue space exposure. No exposure passed the significance threshold corrected for multiple testing of EWAS (Figure 13, Annex I); the exposures most strongly associated with birth weight were dimethyl thiophosphate (DMTP, positive association) and lead and fine particulate matter (negative associations for both exposures).
BMI: Results of the analyses linking the prenatal and postnatal exposome to BMI are presented in Table 8 and Table 9. In the prenatal exposome, exposures that showed a statistically significant association with BMI in the univariate models were active smoking (increased BMI), passive smoking (increased BMI), cotinine (increased BMI), cadmium (increased BMI), and facilities density (decreased BMI); none of these associations passed an FDR corrected p-value of 0.05. The DSA model identified prenatal exposure to cotinine, active and passive maternal smoking and facilities density as contributors to the model but out of these, only active and passive smoking showed statistically significantly associations (p<0.05) with BMI. The p-value for facility density in the DSA model was 0.07. The other indicators of adiposity (waist circumference for abdominal fat, skinfolds for subcutaneous fat, and proportion fat mass from bio-impedence) showed similar results with some small differences in the ranking of exposure variables (not shown). In the postnatal exposome analysis, 34 exposure variables showed significant associations with child BMI in the univariate model and 24 of these had a FDR corrected p-value below 0.05 (table 9, figure 16). These exposures included many POPs: all organochlorines, PBDE153 and some PFAS levels were negatively associated with BMI of the children. Indoor and outdoor air pollution exposures and passive smoking indicators were positively associated with child BMI. Copper and caesium were positively associated with BMI and cobalt and molybdenum negatively. Some dietary intakes (bakery products, fat consumption, breakfast cereals) were negatively associated with BMI. OXBE levels were positively associated with child BMI. Sleep duration was negatively associated with BMI. The DSA model selected indoor PM, copper, HCB, PBDE153 as being significantly associated with BMI of the children.
Spirometry: FEV1% was available for 1,033 (79.4%, ≥70.6% by cohort) children. Values ranged from 60.9 to 139.2 with an average (sd) value of 98.8 (13.2) and large between-cohort variations. In the restricted exposome analysis of prenatal and postnatal exposures, no exposure variable was selected when DSA was applied except for living with cat during childhood; EWAS did not identify any statistically significant exposure-outcome association when correcting for multiple testing. Without correcting for multiple testing in EWAS, 3 prenatal exposure variables were statistically significant at a 5% level: PFOA (decrease), inverse distance to nearest road (increase) and Perfluorononanoate (PFNA, decrease). When adjusting for potential confounding due to co-exposition (i.e. the 9 most significant exposure variables were jointly included in a multivariate linear model), results for the inverse distance to nearest road little varied, while for PFAS and PFNA which were highly correlated (at a 0.61 level), the estimated coefficients lowered and were not statistically significant at a 5% level (Table 10, Figure 18, Annex I). Also, not correcting for multiple testing in EWAS, 6 postnatal exposure variables were statistically significant at a 5% level: MEOHP (decrease), house crowding (decrease), Number of bus public transport mode stops in a 300m buffer around school (decrease), ethyl paraben (decrease), copper (decrease) and Mono-4-methyl-7-oxooctyl phthalate (OXOMINP, decrease). When adjusting for potential confounding due to co-exposition (i.e. the 28 most significant exposure variables were jointly included in a multivariate linear model), coefficients did not vary much, but only ETPA and house crowding remained significant (Table 9, Figure 19, Annex I).
Allergy-related outcomes: The prevalences of the different outcomes were 24, 25, 21 and 10% for itchy rash, rhinitis, eczema and food allergy, respectively. Results of the analyses linking the prenatal exposome to the allergy related outcomes itchy rash last 12 months, eczema ever, rhinitis last 12 months and food allergy ever are presented in Table 11. The EWAS model identified traffic density on nearest road during pregnancy to be positively associated with itchy rash, whereas maternal cotinine levels during pregnancy were inversely associated with itchy rash. However, none of these associations remained statistical significant after the p-value correction. With regard to
rhinitis, PM absorption levels during pregnancy were inversely associated, while inverse distance to nearest road during pregnancy and maternal mono-4-methyl-7-oxooctyl phthalate (OXOMiNP) levels were positively associated with rhinitis. None of the associations remained statistical significant after the p-value correction. None of the exposure variables in the prenatal exposome were significantly associated with eczema. Traffic density on nearest road and the day–evening–night noise level (lden) were inversely associated with food allergy. The inverse distance to nearest road during pregnancy was positively associated with food allergy. None of the associations remained statistical significant after the p-value correction. None of the prenatal exposure remained significantly associated with the allergy-related outcomes (itchy rash last 12 months, eczema ever, rhinitis last 12 months and food allergy ever). However, our findings might indicate a role of exposure in the general external environment such as air pollution, traffic and noise.
Asthma: For childhood asthma, analyses were performed on around 1300 children from whom questionnaire data was available. The prevalence of parent reported ever asthma at age 5 ranged from 3.7% (SAB cohort) to 20.2% (EDEN cohort). Results of the EWAS analyses linking the prenatal exposome to asthma are presented in Figure 20. None of the exposure variables in the prenatal exposome were significantly associated with asthma.
Neurodevelopment: Several components of neurodevelopment have been assessed in HELIX children: cognition (assessed from Raven test), internalizing and externalizing disorders, both assessed using CBCL tool. Figure 21 shows the relations between Prenatal Exposome (left-hand panel) and child CBCL internalizing problems. Passive smoking during pregnancy is strongly associated with more problems. In the right-hand panel of Figure 21, we show the association between the Postnatal Exposome and internalizing problems. Sleep duration reduces such problems but PCBs also were associated with a reduction in internalizing problems. Figure 22 reports associations with externalizing problems. The Prenatal exposome shows strong associations with active and passive smoking during pregnancy and, at postnatal periods, it shows that healthy diet is protectively associated with child problems but readymade foods, sweets and indoor air pollution are associated with more eternalizing problems.
The prenatal exposome did not show any strong association with cognition based on Raven test (Figure 23), with facility density and mercury exposure being positively associated with the test. The postnatal exposome showed organic food positively associated with the raven scores and fast food and house crowding showed associations in the opposite direction (Figure 23, right-hand panel).
Blood pressure: Results of the analyses linking the prenatal and postnatal exposomes to child blood pressure are presented in Table 12 and Table 13. After p-value correction, significant decrease in systolic BP was observed with markers of the built environment during pregnancy (e.g. facilities density) and with child concentration of some organochlorine compounds (e.g. DDE). Other findings (p<0.05) that did not remain significant after p-value correction could be mentioned: 1) prenatal exposure to bisphenol-A was associated with an increase in both systolic and diastolic BP, 2) low and high fish consumption during pregnancy were associated with higher systolic BP, 3) outdoor temperature and vitamin-D UV dose the day of BP measurement were associated with a decrease in diastolic BP, 4) child concentration of copper and exposure to benzene was associated with an increase in diastolic, 5) Postnatal exposure to benzene was associated with increase in systolic BP, and 6) child concentration of phthalates were associated with a decrease in both systolic and diastolic BP. Results from the DSA selections were concordant with EWAS analyses, with a higher parsimony regarding the selection of exposure within family. Also, DSA selected additional exposures, which taken individually were not associated with blood pressure (i.e. maternal cotinine and child PFOA confounded by a co-exposure). When the prenatal and postnatal exposures selected by DSA were jointly included in a multivariate linear model (Table 14), coefficients differed slightly from those of the univariate models.
Conclusions
HELIX enabled important methodological progress in the study of the Exposome-health relations, and provided one of the first real implementation of an analysis of the potential effects of the early-life Exposome on several components of children health. From a methodological point of view, the EWAS approach used in some former Exposome-health studies was shown to heavily suffer from lack of control for false positive rate. Other more efficient statistical approaches were identified, some of which allowing to characterize interactions between exposure with no cost on statistical power. Our study is one of the largest simultaneously considering over 70 environmental exposures for effects on several components of children health. Compared to former (repeated) single exposure studies, our approach allowed making all tests (usually done in successive studies) explicit, correcting for confounding by co-exposures; we also made attempts in some analyses to correct for exposure misclassification and considering interactions between exposures.
Associations of BMI with postnatal (but not prenatal) levels of Persistent Organic Pollutants need to be considered with caution given the potential for reverse causation. The prospective nature of HELIX is a clear strength given the increased potential for such bias in cross-sectional analyses. By simultaneously testing a large number of exposure factors, the Exposome approach allows to discard confounding by co-exposures and explicitly account for multiple testing, paving the way for a better characterization of the overall influence of environmental factors on human health.
WP6 - Environmental Burden of Childhood Disease and Health App
WP6 aims to estimate the burden of common childhood diseases (obesity, asthma and neurodevelopment) that may be attributed to multiple environmental exposures in Europe. Specific objectives for the full project were:
▪ To construct scenarios for a health impact assessment;
▪ To obtain exposure estimates;
▪ To obtain exposure response data;
▪ To obtain burden of disease estimates;
WP6 first developed an evaluation of the environmental burden of childhood disease in European Union (28 countries). In addition, scenarios for health impact assessment (HIA) have been quantify on active transportation in children, that will integrate multiple exposures (e.g. air pollution, physical activity, traffic incidents) and outcomes (e.g. mortality, morbidity, etc.). The HIA was based on the exposure and outcome data obtained from WP 1-5. In addition, the HIA work includes the design and development of a quantitative model to estimate the risks and benefits of the active transportation in children.
Environmental burden of childhood disease
In this study we aim to estimate the burden of childhood disease due to environmental risk factors in the European Union (EU) of the 28 countries, describing the impact of six environmental exposures (particulate matter less than 10 micrometer of diameter (PM10) and less than 2.5 micrometer of diameter (PM2.5) ozone, secondhand smoke, dampness, lead and formaldehyde), identifying priorities in environmental health policies for childhood in Europe, and highlighting research and risk management necessities.
Selection of environmental risks and health outcomes
This burden of disease was focused on environmental risk factors in children. Metabolic and behavioral risk factors (e.g. sedentary, nutrition, active smoking), and infectious diseases, were excluded from the assessment. The selection of environmental risks and health outcomes were based on the following criteria: evidence for a causal relationship between exposure to the environmental risk factor and the health effect (based on meta-analyses, World Health Organization (WHO) guidelines, or previous risk assessments), independent health effects between the risk factors, availability of exposure data at national level, and availability of baseline health statistics at national level.
Data Collection
Our study considered the population between 0 to 18 years old of the 28 European Union Countries (EU28) (Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxemburg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden and United Kingdom). Population data by country and age were collected from Eurostat database and Institute for Health Metrics and Evaluation (IHME).
Health data was collected from the Institute for Health Metrics and Evaluation (IHME) and the World Health Organization for asthma, mild mental retardation, otitis media, lower respiratory tract infections and infant mortality. From scientific papers and reports, data for cough and low respiratory tract symptoms were collected. Exposure data were collected from IHME for lead, PM2.5 ozone and secondhand smoke, and Environmental and Health Information System (ENHIS) for dampness.
Environmental Burden of Disease
The environmental burden of disease analysis was performed following the Comparative Risk Assessment Approach proposed by the World Health Organization, and the global burden of disease project. The environmental burden of childhood disease was estimated for exposures above defined thresholds, if any, based on a counterfactual exposure distribution that would result in the lowest population risk. The feasibility of reaching the counterfactual exposure levels in practice was not assessed in this proposal. The burden of disease was estimated using the exposure data and a relative risk (RR) derived from epidemiological studies to estimate the population attributable fraction (PAF). This analysis was applied to each exposure-outcome pair. The PAF is defined as the proportional reduction in disease or death that would occur if exposure to the risk factor were reduced to the counterfactual. The burden of disease was calculated using disability weights (DW) and estimates of duration (L) of each outcome.
See below the explanation using three general formulas:
(1) PAF = f x (RR-1 ) / f x (RR-1)+1
(2) AP = PAF x P
(3) DALY = AI x DW x L
Where: PAF: Population attributable fraction; f: Fraction of population exposed; RR: Relative risk; AP: Attributable prevalence; P: Background prevalence; DALY: disability-adjusted life year; DW: Disability weights; L: Duration of condition.
The disability weights used in this analysis were those proposed by the global burden of disease project or used in previous environmental burden of disease estimations. For some cases an approach based on the unit of risk (UR) was used. The UR estimated the absolute number of cases that are to be expected at a certain exposure, and then was transformed to DALYs using disability weights and the duration of the condition. Burden from lead and mental retardation, PM2.5 and low respiratory infections, secondhand smoke and low respiratory infections and otitis media were obtained from the global burden of disease project 2015.
Environmental Burden of Disease in the EU of 28 countries.
Seven different exposures where identified under the inclusion criteria, associated to six different health outcomes. We estimated that in the population aged below 18 years of the EU28, the seven environmental exposures (lead, PM10, PM2.5 ozone, secondhand smoke, dampness and formaldehyde), are responsible for 210777 disability adjusted life years (DALYs) annually. Fifty nine percent of these DALYs were attributable to particulate matter (PM10 and PM2.5) exposure, 20% to secondhand smoke, 11% to ozone, 6% to dampness, 3% to lead, and 0.2% related to formaldehyde.
Particulate matter less than 10 micrometer of diameter (PM10) and less than 2.5 micrometer of diameter (PM2.5).
PM10 is associated with infant mortality (< 1 year old)(WHO/Europe 2013) and asthma (5-18 years old). Of these, infant mortality was associated with the major burden (93147 DALYs annually), followed by asthma (13904 DALYs annually). PM2.5 is associated with low respiratory infections (< 18 years old) and was estimated to produce 134032 DALYs annually.
Secondhand smoke.
Secondhand smoke is associated with asthma (< 14 years old), low respiratory infections (< 5 years old) and otitis media (< 5 years old). Of these, asthma was the disease with the mayor burden, resulting in 20880 DALYs annually, followed by low respiratory infections with 9728 DALYs annually, and Otitis media with 2062 DALYs annually.
Ozone.
Ozone is associated with low respiratory symptoms (including cough)(5-14 years old). Cough days related with ozone was estimated to result in 10057 DALYs annually. Other days with low respiratory symptoms related to ozone were estimated to result in 14122 DALYs annually.
Dampness.
Dampness is associated with asthma in children less than 14 years old, this disease was estimated to result in 12954 DALYs annually.
Lead.
Lead exposure is associated with mild mental retardation (< 14 years old), this disease was estimated to result in 6216 DALYs annually.
Formaldehyde.
Formaldehyde is associated with asthma in children less than 3 years old, and resulted in 33 DALYs annually.
Sensitivity analysis.
Sensitivity analysis for PM10 and asthma assuming a counterfactual of 1.9 μg/m3 resulted in 18681 DALYs, and assuming counterfactual of 20 μg/m3 resulted in 3885 DALYs. We also performed a sensitivity analysis using a different exposure-response function between PM10 and asthma from new meta-analysis; in this analysis was estimated 45098 DALYs. The sensitivity analysis of PM10 and infant mortality assuming a counterfactual of 1.9 μg/m3 resulted in 124794 DALYs, and assuming counterfactual of 20 μg/m3 resulted in 30499 DALYs. Sensitivity analysis for secondhand smoke assuming the minimum percentage of secondhand smoke reported in the European Union 28, resulted in 12848 DALYs for asthma, 6867 DALYs for low respiratory infections, and 1168 DALYs for otitis media. Sensitivity analysis for dampness and asthma, using mould (instead of dampness) as an exposure resulted in 11470 DALYs. Finally the sensitivity analysis for formaldehyde and asthma using a 60 μg/m3 as a threshold resulted in 1667 DALYs.
Development of the DigiBIB APP: linking environmental exposures and electronic health records.
Using learning from earlier work-packages we have developed a multi-functional prototype smartphone APP which can track users’ locations and record research calibre real time assessments of health and wellbeing, with the potential to calculate personalised exposure to air pollution and other exposures. The APP also allows users to access their own electronic health record, and will allow dyadic communication between health professionals and participants (for example, along medical professionals to gain contextual information about levels of pollution users are experiencing, and allowing users to report health concerns such as wheezing and asthma). The APP is currently being piloted within the Born in Bradford cohort study where we will explore acceptability and usage amongst different socio-economic and ethnic groups.
Conclusions
Our results show a large impact of environmental exposures in child health across Europe (Figure 24, Annex I). This study found that the environmental risk factor for child health in the EU28 with the largest impact was air pollution (PM10, PM2.5 and ozone) exposure, representing more than two thirds of the environmental burden of disease of the seven exposures combined. This is similar to a previous burden of disease study in six countries of Europe. Secondhand smoke also showed a large impact, resulting in 20% of the environmental burden of disease in European children.
The assessment of environmental burden of disease estimates offers the opportunity to identify priorities and solutions for policy and research. These priorities and solutions are presented in this study as recommendations for authorities, public health specialist and researchers, and are as follows. First, this study found a real need to create and implement effective policies to reduce children exposure to environmental risk factors across Europe, with special attention to major risk factors such as air pollution and secondhand smoke. Second, there is a need to create common European databases, which collect and harmonize exposure data for (old and new) environmental risk factors, especially for children. Third, the need for epidemiological studies on multiple environmental risk factors, with special attention to provide dose-response functions, with harmonized exposure and outcome definitions. Currently, a number of European research projects focus on the “exposome” concept as HELIX, EXPOsOMICs and HEALS (HEALS 2017), and some of them collected relevant information on multiple exposures and outcomes in children. These projects should be seen as a good examples to start the filling some of these gaps (Table 17, Annex I).
Potential Impact:
Key messages and implications
Novel Tools for Individual Exposures
▪ Helix has demonstrated that it is possible to perform a harmonized, though extensive, fieldwork using the same protocols for sample collection and clinical examination in six European mother-child cohorts.
▪ Children across Europe are exposed to a wide range of environmental chemicals in fetal life and childhood, and the exposure varies substantially between countries and by compounds.
▪ The concentrations found in the maternal samples were in general higher than concentrations measured in the child samples.
▪ For most of the persistent compounds the correlation between maternal and child levels was high, while considerably lower correlations were observed for most non-persistent compounds.
▪ There is a substantial variability in exposure during pregnancy and over a year in school-age children for the majority of the non-persistent compounds.
Novel Tools for Outdoor Exposures
▪ HELIX developed and applied standardized exposure assessments for urban exposures including built environment indicators, air pollution, noise, green space and temperature and successfully applied these to different study areas.
▪ Urban exposures vary substantially within and between children and pregnant women and within and between cities.
▪ Generally, correlations between the different urban exposures are weak to moderate.
▪ A considerable proportion of children and pregnant women are exposed to levels of urban exposures that are above international guidelines.
Integrating Multiple Exposures and Uncertainties
▪ The exposome is high-dimensional, as it cannot be reduced to a small set of principal components.
▪ Although correlations within the same exposure family can be high, correlations between exposures from different families were low. This supports that epidemiological studies focusing on a reduced set of exposures may not be confounded by having omitted exposures from other families.
▪ Clustering of exposures may not be adequate for multicentre analyses on the effects of the exposome on health outcomes as location is a strong determinant of one’s personal exposome.
▪ PBPK models can be used to integrate more individual information (e.g. breastfeeding) together with biomarker measurements in order to rebuild realistic exposure scenarios.
▪ For most non-persistent chemicals, multiple pools of multiple urines would be needed to obtain excellent reliability in exposure assessment; for example, 4 pools of 15-20 urines each would be needed to get a reliable estimate over the entire pregnancy. This is important for the design of future biomarker studies and for measurement-error corrections in HELIX.
▪ The variability in environmental outdoor exposures was mainly determined by the activity patterns of participants’ daily life rather than by individual or city characteristics. These results confirm the need of using personal exposure assessment methods for outdoor exposures in epidemiological studies and the importance of having access to accurate, inexpensive and not burdensome personal exposure tools.
▪ Future exposome studies should continue refining exposure assessment through repeated collection of biospecimens and personal dosimeters in even larger populations.
Integrating Molecular Exposure Signatures
▪ In the HELIX project we have successfully generated a unique molecular profile data resource for exposome research comprising of urinary and serum metabolomics, plasma proteomics, blood cell DNA methylation, transcriptomics and miRNAdata for 874 children in the HELIX subcohort, with up to 1198 children’s samples analysed for individual omics platforms.
▪ An EXposure Wide Association Study (ExWAS)conducted with metabolomics, proteomics and the child-matched full exposome dataset from HELIX identified several significant clusters of associations.
▪ Serum polyunsaturated glycerophospholipids were associated with exposure to heavy metals and PFASs, also urinarytrimethylamine-N-oxide with exposure to heavy metals and urinary hippurate and proline betaine with OPs. These signatures are likely to represent common routes of exposure, e.g. fish and seafood as a common source of polyunsaturated fatty acids and metals.
▪ We also found associations between OCs and PBDEs with plasma adipokines, which can reflect a link with fat mass, e.g. fat mass as a driver of adipokine expression and storage of lipophilic chemicals.
▪ We have defined the major determinants of the metabolome in European children and identified a novel link between threonine catabolism and BMI.
Linking the Exposome to Child Health
▪ Simulation studies indicated that in the presence of correlation between exposure factors, Exposome wide association studies, even correcting for multiple testing, are expected to suffer from a high rate of false positives.
▪ Other statistical approaches, such as the Deletion/Substitution/Addition (DSA) or Elastic Net algorithm, have a lower false detection rate, at a cost of a decreased sensitivity for a given sample size. The efficiency of several approaches aiming at identifying statistical interactions between exposure factors has also been characterized.
▪ Studies linking the Exposome to various component of children health (birth weight, postnatal growth, respiratory health and allergy, blood pressure, neurodevelopment) have been conducted in 1300 children, as well as a study linking the urban Exposome to birth weight in 30,000 children.
▪ The Exposome health associations observed in Helix include a relation between lead and green space exposure and birth weight; relations between active and passive smoking (prenatally), indoor particulate matter, copper (postnatally) and body mass index; facility density (prenatally), DDE and HCB (postnatally) and systolic blood pressure; passive smoking (prenatally), PCB and sleep (postnatally) and internalizing problems.
▪ By simultaneously testing a large number of exposure factors, the Exposome approach allows to discard confounding by co-exposures and explicitly account for multiple testing, paving the way for a better characterization of the overall influence of environmental factors on human health.
Environmental Burden of Childhood Disease
▪ HELIX has estimated the childhood environmental burden of disease in Europe, highlighting the relevance of environmental factors in children's health across the European Union.
▪ HELIX has developed new evidence that could be integrated into future health impact assessments to estimate more comprehensively the environmental burden in childhood health and to assess future policy interventions.
▪ HELIX has also highlighted the need for more evidence on the Exposome that can be translated into evidence-based policymaking to protect and improve public health.
Use and Dissemination of results
HELIX set out to produce more robust evidence base on links between exposome and human health and well-being. The HELIX project produces a lot of data and hence results on exposome research. As a reminder the objectives of the HELIX project are reiterated as follows:
• To measure a range of chemical and physical environmental hazards in food, consumer products, water, air, noise, and the built environment, in pre- and post-natal early-life periods;
• To define multiple exposure patterns and individual exposure variability (temporal, behavioural, toxicokinetic);
• To quantify uncertainty in the exposure estimates;
• To determine molecular profiles and biological pathways associated with multiple exposures;
• To obtain exposure-response estimates for multiple exposures and child health outcomes;
• To estimate the burden of childhood disease in Europe due to multiple environmental exposures;
• To strengthen the knowledge base for European policy.
Through these objectives and associated methods, a significant amount of data has been and will be produced and interpreted. This will lead to conclusions that can be used in health/environment/research policies. The results can help us to plan ways of improving public health. Moreover, the results can be useful for policy makers in the domain of environment and health and even in the public health domain. The results of the project will also point to further open questions which can lead to more research.
The project has delivered a wide range of results. These are produced in different formats: Scientific reports; scientific articles; guidelines and recommendations for stakeholders or professional practitioners; dissemination at a range of meetings.
The products in short are:
O Scientific article
O Newsletters
O Brochure
O Press releases
O News articles
O Conference/meeting attendance (scientific, advisory board, policy)
O Summary future research needed
O Section implementation in paper
O Social media/media
Each of these formats have their own value in dissemination of the results of the project.
The results in HELIX have been produced in seven work packages (WP). Each WP has produced its own reports with results, as well as articles in peer reviewed journals. There are different categories of stakeholder groups that are addressed in HELIX.
In order to ensure stakeholder involvement, we have used the stakeholder network of the HELIX project. This task involved establishing and maintaining structural and ad hoc contacts with policy makers, regulators, research networks and other stakeholders. At the start of the project a Stakeholders Forum was suggested across the different Exposome projects. This has not been realized yet, but a proposal for a network has been filed as a COST Action at the EU. An inventory of stakeholders' needs is the starting point of stakeholder involvement. There is a need in Europe for an Exposome toolbox as an important source of information in the decision making process in the domain of environment and health.
These stakeholder groups have been identified within the consortium by the partners, in collaboration with stakeholders attending the different meetings within HELIX.
Scientific reports: Results based on reports of the different WPs aim at scientists and policymakers. These reports have been or will be distributed by the work packages and disseminated via the HELIX website. Other stakeholders can also access these reports.
Scientific articles: The scientific articles have been produced by the members in the HELIX consortium (Table 18, Annex I). Most articles have been published in peer reviewed journals. Even more articles have been submitted and are currently being drafted (Table 19, Annex I). The abstracts of these articles are provided in Deliverable D7.11
General public: The general public has not been a prominent stakeholder so far for the consortium to disseminate the results, however, the partners have presented the project at public events throughout the project. The topic of HELIX is complicated and not yet developed into full strength in order to disseminate the results in an easy to understand manner.
Throughout the project, partners have contributed to a significant number of (newspaper) interviews and video reports in the popular media. The general dissemination about the concept of the HELIX project to the general public has taken place in different countries. In addition, video messages were produced and disseminated via the project website (http://www.projecthelix.eu/).
Webinars: The project established a collaboration with the Exposure Science and the Exposome Webinar Series initiative of the National Institute of Environmental Health Sciences (NIEHS, USA). The first HELIX Webinar was recorded in October 2014, followed by a joint Stakeholder Webinar with the EXPOsOMICS project. The last combined webinar was held in December 2015 as joint session with NIEHS and Exposomics. We intend to keep collaborating on this type of engagement, where possible with other initiatives.
Stakeholder interaction: The interaction with stakeholders outside the consortium has been analysed in HELIX Dissemination Strategy (D7.10). Furthermore, the most favorable engagement with stakeholders has been described.
The results of the research of HELIX are important for different stakeholders. Results of research projects do have more value if they are translated into practical and understandable recommendations. This process is not always easy. Results from research always raise more questions. However, each result will point in a certain direction. If some health benefit can be added to a result out of the HELIX research it should be mentioned.
The presented results of the different studies within HELIX all have some value for implementation. Science produces a lot of results which can lead to more scientific questions. But there are always results that can be used to improve our world. Sometimes these results are still very indicative. Sometimes results are not causal to explain mechanisms. But always there are possible conclusions that cause no harm to take action upon.
A result can be descriptive and confirm common knowledge. There are some items that have been studied that are a proxy for a range of factors that cannot be described.
The results of the research of HELIX are important for different stakeholders. Results of research projects do have more value if they are translated into practical and understandable recommendations. This process is not always easy. Results from research always raise more questions. However, each result will point in a certain direction. If some health benefit can be added to a result out of the HELIX research it should be mentioned.
To enhance the potential impact of the project and improve dissemination at all levels, HELIX has released a set of recommendations. These are formulated based on the scientific field work and on literature analysis.
Accessible Data Inventory
As part of the remit, HELIX has constructed the HELIX Data Warehouse, containing HELIX Foreground data (i.e. data generated by the HELIX study) and participating cohort Background data (i.e. data originally generated by the individual cohorts). The data is available for research purposes to researchers outside the Consortium for work on specified manuscripts. A clear overview of the inventory and included variables is available from www.projecthelix.eu/index.php/es/data-inventory. Interested parties find here the request protocol has been established to guarantee adherence to cohort data access rights, data protection regulations, and ethics approvals relevant to the cohorts participating in HELIX.
Researchers external to the HELIX Consortium who have an interest in using data held in the HELIX warehouse for research purposes can apply for access to data for a specific manuscript at the time. The analysis of these data will be considered outside the remit of the HELIX project. The applicant has to be affiliated with an institution with competence in conducting research projects and which has agreed to be responsible for the conduct of the proposed manuscript. Junior researchers must have a scientific supervisor belonging to such an institution. All proposed manuscripts must have a principal investigator with scientific responsibility for the project. For each accepted proposal, a data transfer agreement (DTA) will be signed between the cohorts participating in HELIX, and the receiving institution. A detailed description of the Data Warehouse can be found in Deliverable 5.4 submitted to the EC following the official end of the project.
Further procedures to for the request of biological materials available from the participating cohorts will be finalized and published during the course of 2018.
Policy implications
The HELIX project has produced different products. The project is finished at a stage were most results of individual studies are analysed and made ready for publication. The final scientific symposium produced a limited amount of suggestions of policy and societal implications of the results of the studies.
The complicated exposure scenarios can lead to large exposure mis-classification problems. However, the combination of PBPK modelling and realistic exposure data can contribute to rebuild realistic exposure scenarios and reduce uncertainty. This approach can lead to the identification of very sensitive individuals. Results have shown this for PBPK modelling of PFOA and PFOS.
Another implication of the studies shows that intervention early life exposure can contribute to decrease comorbidities in adulthood. Therefore we need standards based on vulnerable time periods.
There are several individual studies that have analysed health risks related to multiple exposures in a novel systematic way and these show possible implications for health. These are related e.g. with environmental exposures and child blood pressure, endocrine disruptors and lung functions, organic compounds and ADHD or social competence, environmental exposures and obesity.
These findings show that we should develop:
▪ Governmental policies to reduce environmental risk factors
▪ Early prevention in children that should be effective (children with elevated BP are more at risk to be hypertensive in adulthood)
▪ Measures to improve quality of life and decrease health costs
Ultimately, results would help us to guide public health efforts by allowing us to intervene on those chemical agents or urban exposures that are most likely to be associated with childhood diseases. Results would thus help prioritisation of interventions.
Evidence-based knowledge translation
The scientific evidence created by HELIX will help to disentangle the relationship and impact of multiple exposures and health outcomes. To translate this scientific evidence into decision-making processes will require the development of tools like Health Impact Assessment. During the HELIX project, a health impact assessment approach was used to estimate the health impacts of seven environmental exposures in the European Union (EU28). This exercise helped to identify the main priorities (based on evidence) between these seven environmental exposures individually in the EU28. Applying a health impact assessment approach using the Exposome concept instead of using individual exposures will require more scientific evidence, especially from epidemiological studies, and the development of further approaches to translate the Exposome concept to a more integrated unit.
Until now the evidence around Exposome is not yet integrated into a simple and harmonize unit, in order to be translated to policies and design interventions. For that reason, in the future, the development of a simple approach to translating the Exposome concept into policies and interventions will be required. A possible way to tackle this would be the development of an index, which integrates different dimensions and characteristics of the Exposome and could be used to communicate the health risks and benefits of multiple exposures in a single unit. The “Exposome Index” could be an easy tool to identify where and when a policy or intervention will be required, will also help to prioritize interventions, compare populations, regions, and develop trends. Finally the “Exposome Index” could also help to assess the efficacy of policies and interventions, based on their performance to improve health determinants and health outcomes. Further research will be needed to develop the “Exposome Index” to the available evidence around the Exposome concept; the potential to develop an “Exposome Index” could simplify the translation of the Exposome concept (and evidence) into policy.
Continuation
During the course of the project, the objective was to cluster coordination activities between the HELIX WPs. The focus of this activity concerned HELIX WP7 (Dissemination and Engagement) and WP8 (Management) intent to exploit outcomes of the project most efficiently.
Amongst other, this concerned the development of a dissemination strategy that included the identification of stakeholders, identification of appropriate dissemination frameworks and methods, with exchange between and coordination of, appropriate initiatives. This is outlined in Deliverable 7.10 Final Dissemination Strategy.
These recommendations for future dissemination are depending on the progress of the analysis of the collected data in HELIX. The plan is to continue to publish scientific papers after the official end of the project. It will take at least another year before most of the papers are published. All the researchers within the consortium are going to fill in the templates about the results of their scientific efforts.
The compilation of all of the templates will be added to the website of the project. Thus, the information will be available for the stakeholders. With help of the provided templates it will be more convenient to have an overview of all the results. In addition, the researchers are asked to add any created visuals to bring across their message, by using Infographics, shared presentations / videos on YouTube or SlideShare.
The HELIX website (http://www.projecthelix.eu/) will all add the connection to future studies, new technologies etc. based on the samples collected during the project (continuation process). The website and HELIX social media accounts - twitter (@greenhealth4eu), Facebook and LinkedIn - will continue to be active and function as reference sources and as tools for viral messaging.
In the long-run, exploitation of the HELIX results and the knowledge acquired during the HELIX project will be achieved via involvement of partners in other international projects, including those funded by the EU (Table 20, Annex I). HELIX partners are already involved in mayor H2020 projects including LifeCycle (Early-life stressors and LifeCycle health) and HMB4EU (Science and policy for a healthy future), as well as the newly proposed STOP (Science and Technology in childhood Obesity Policy). Further exploration of the gaps in exposome research and use of HELIX data is expected to lead to a significant number of new proposals in the near future. Furthermore, the involvement of partners in policy networks at the national and international level, including different Directorates of the EU, Member States’ health ministries and their commissions, and regional health authorities, will facilitate the awareness of HELIX results and their possible use for the development and implementation of policies related to public health and environment.
Plan for continued dissemination of project results
Webinars
The project established a collaboration with the Exposure Science and the Exposome Webinar Series initiative of the National Institute of Environmental Health Sciences (NIEHS, USA). The first HELIX Webinar was recorded in October 2014, followed by a joint Stakeholder Webinar with the EXPOsOMICS project. The last combined webinar was held in December 2015 as joint session with NIEHS and Exposomics. We intend to keep collaborating on this type of engagement, where possible with other initiatives.
Brief
A short brief on what HELIX does and what ‘exposome’is has been produced for dissemination purposes (Table 21, Annex I). This brief is available on the website and will be used in communication about ‘exposome’.
Newsletter
Electronic newsletters send through the earlier mentioned Mailchimp system have been published at regular intervals during the project. The news alerts aim at alerting users to key developments and headline results, events, and personnel involved in the project, and to provide links to other related activities (e.g. parallel studies). It also provides brief commentaries on the implications of these developments for policy, and invites similar commentaries and reviews from users or other researchers. We will continue this effort in order to update stakeholders of ongoing developments and publications. This resource also allows for new stakeholder to sign up as they see appropriate, providing for the dynamic changes of people entering and leaving the field of interest.
Workshops
Efforts are made to find financial and time resources to organise HELIX-exposome focussed meetings beyond 2017. The earlier mentioned COST Action, is a potential way to organise workshops to a network of exposome researchers. Also workshops on Exposome can be organised in different scientific conferences in 2018 e.g. PPTOX VI and ISEE 2018.
Policy impact
The department of Policy and Global Development is the core element in ISGlobal's knowledge transfer strategy. The department's dual function as a think tank and a catalyst for ideas and action embodies the institute's strategy of studying real world problems to effect change. The HELIX management is collaborating with this department in its efforts to translate the scientific results into clear recommendations for policy advisors and decision makers. Several partners within the HELIX consortium have discussed the topic of policy transfer of Exposome research results. This Policy Core Group has made a work plan on how to deal with science policy transfer. The set-up of the final scientific HELIX conference has been built upon insights of the workplan.
Further resources
Publication of materials for the non-specialist audience is a tool within the project. Moreover, the use of illustrations, tools and further media developed within the project will be actively encouraged. This can be incorporated in the dissemination of the templates as they are proposed in deliverable 7.11 – the report on awareness and societal implications of HELIX research and development.
A stand-alone document including the project description, abstract, the main policy relevant questions that were addressed, and first results, will be prepared in lay language.
List of Websites:
www.projecthelix.eu
The HELIX Project aimed to characterise the early-life exposome by assessing exposure to many environmental hazards during pregnancy and childhood, and linking these to children’s molecular omics signatures and to their cardiometabolic health, respiratory and immune health, and neurodevelopment (Vrijheid et al 2014). HELIX has successfully generated geospatial exposure estimates for outdoor air pollutants, noise, green space, meteorological and urban environment factors in around 30,000 mother-child pairs from 6 European birth cohorts during pregnancy and childhood. In 1301 cohort subjects HELIX also generated estimates for prenatal and postnatal exposure to chemical pollutants using highly sensitive biomonitoring methods. In total, over 200 separate exposure variables were measured. Complete molecular profile data sets comprising of urinary and serum metabolomics, plasma proteomics, blood cell DNA methylation, transcriptomics and miRNA data were generated for the same children. Harmonized child health outcome data were collected by applying common fieldwork protocols in the 6 countries. In doing this, HELIX has demonstrated that it is possible to build an early life exposome database with completely comparable biomonitoring data, geospatial data, child health outcome data, and omics signatures in the same subjects using a large-scale prospective framework.
First results focus on describing the Exposome, its correlations and its determinants, and highlight that: 1) exposure to most of the chemicals measured was abundant, with many being detected in over 90% of participants; 2) a considerable proportion of pregnant women and children are exposed to levels of urban exposures above international guidelines; 3) the exposome showed considerable variability across Europe for both urban and chemical exposures, confirming that location is a strong determinant of one’s personal exposome; 4) correlations within the same exposure family can be high, but correlations between exposures from different families were low, supporting the notion that epidemiological studies focusing on a single family of exposures may not be confounded by exposures from other families; 5) the exposome proved to be highly-dimensional and cannot be reduced to a small set of principal components; and 6) panel studies quantifying within-person variability in biomarker and outdoor exposures conclude that repeat biospecimen collections and personal dosimeters are important tools to refine exposome assessments and should form an integral part of future exposome studies.
An important challenge in associating the exposome with health is the simultaneous consideration of many correlated exposures. HELIX developed a statistical approach to evaluate Exposome-health associations in the light of complex correlation patterns, recognising that statistical techniques are often limited in their ability to efficiently differentiate true predictors from correlated covariates (Agier et al 2016, Barrera et al 2017). Within this framework, the systematic evaluation of child health risks related to multiple exposures will ultimately help guide public health efforts by allowing us to intervene on those chemical agents or urban and lifestyle exposures that are most likely to be associated with child health; indeed, preliminary results indicate a number of suspect exposures, including lead and reduced green space in relation to birth weight and maternal smoking in relation to child obesity. Also, HELIX is developing a catalogue of omics signatures associated with multiple environmental exposures, in order to better understand molecular mechanisms and early signs of damage, and potentially identify biomarkers; first results indicate that specific exposures tend to be associated with specific molecular signatures, e.g. methylation signatures for smoking, and metabolomics signatures for heavy metals and pesticides.
The database and scientific evidence created by HELIX will help to disentangle the relationship between multiple environmental exposures and health outcomes. To translate this scientific evidence into decision-making processes will require the development of tools like Health Impact Assessment (HIA). During the HELIX project, the HIA approach was used to estimate the child health impacts of seven individual environmental exposures in the EU, identifying air pollution as having the main impact.
By putting together over 200 exposures and omics and health data in a large, prospective early life framework, HELIX confirms that many environmental factors constitute a hazard for children in Europe. HELIX paves the way for a better characterization of the overall influence of environmental factors on human health.
Project Context and Objectives:
Background
The "exposome" concept encompasses the totality of non-genetic exposures from conception throughout the life course, complementing the genome. The exposome concept carries the expectation that the use of holistic and data-driven approaches, similar to those pioneered in the genomics fields, can result in advances in our understanding of the complex environmental component of disease aetiology. The exposome has been delineated to include three overlapping and complementary domains: 1) a general external domain including macro-level factors such as climate, urban environment and societal factors; 2) an individual external domain including agents such as environmental pollutants, tobacco smoke, diet and physical activity; and 3) a specific internal domain including gene expression, inflammation, and metabolism, often assessed through high-throughput molecular omics methodologies such as transcriptomics, proteomics and metabolomics.
The developing fetus, infant and child may be especially vulnerable to effects of environmental exposures since these are periods of rapidly growing and developing organs, immature metabolism, and the received dose relative to bodyweight may be greater than later in life. In utero and early life exposure to environmental stressors during critical windows can disrupt developmental processes and under The Developmental Origin of Health and Disease hypothesis such effects may permanently alter body structure, metabolism and physiology, and be expected to have a lifetime impact. Therefore the “early life exposome” is key, both as a starting point to develop a lifetime exposome and due the heightened impact of the exposome at this time. There is now moderate to good evidence for the effects of prenatal exposure to environmental contaminants, including air pollutants, polychlorinated biphenyls (PCBs), lead, mercury and organophosphate pesticides on fetal growth, neurological development, and the respiratory and immune systems. Evidence is also growing for effects on childhood growth, obesity, and metabolic signaling. At the same time it is clear that, up to now, the environment and child health field has almost uniquely focused on single exposure-health effect relationships. Only a few studies so far simultaneously considered more than a couple of families of compounds, focusing health outcomes such as birth weight, fecundity or type II diabetes mellitus.
Objectives
The HELIX project aims to measure and describe multiple environmental exposures from the three different exposome domains during early life (pregnancy and childhood) and associate these with omics markers and child health outcomes. The background and rationale of the HELIX project have been fully described (Vrijheid et al 2014). The objectives of HELIX were defined as follows:
Step 1: Measuring the external exposome
- To obtain estimates of exposure to persistent and non-persistent pollutants in food, consumer products, water and indoor air, during pregnancy and in childhood.
- To obtain estimates of chemical and physical exposures in the outdoor environment during pregnancy and in childhood: ambient air pollution, ambient noise, ultraviolet (UV) radiation, temperature, and built environment/green space.
Step 2: Integrating the external and internal exposome:
- To define multiple exposure patterns in the individual and outdoor environment, describe their predictors, and describe uncertainties and variability in the exposures assessed.
- To measure molecular signatures associated with multiple environmental exposures through analysis of profiles of metabolites, proteins, transcripts, and DNA methylation in biological samples from the children in the cohorts.
Step 3: Impact of the early-life exposome on child health
- To develop a novel multi-step statistical approach for the analysis of the association of patterns of multiple and combined exposures and child health outcomes.
- To provide exposure-response estimates for the association between multiple and combined exposures, and child health focusing on foetal and childhood growth and obesity, neurodevelopment, and respiratory health.
- To estimate the burden of common childhood diseases that may be attributed to multiple environmental exposures in Europe.
- To strengthen the knowledge base for European policy in the area of child and environmental health by engaging with, and effectively disseminating HELIX knowledge to, stakeholders including those responsible for risk management and mitigation and prevention strategies.
General Study Design
The HELIX study represents a collaborative project across six established and ongoing longitudinal population-based birth cohort studies in Europe: the Born in Bradford (BiB) study in the UK, the Étude des Déterminants pré et postnatals du développement et de la santé de l’Enfant (EDEN) study in France, the INfancia y Medio Ambiente (INMA) cohort in Spain, the Kaunus cohort (KANC) in Lithuania, the Norwegian Mother and Child Cohort Study (MoBa), and the RHEA Mother Child Cohort study in Crete, Greece. These cohorts were selected for participation in the HELIX project because: a) they could provide substantial existing longitudinal data from early pregnancy through childhood, b) they could follow-up children at similar ages, c) they could integrate questionnaires, biosampling and clinical examinations using common HELIX protocols, and d) they offered heterogeneity in terms of exposures and population characteristics.
Pregnant women in the original cohorts were recruited between 1999 and 2010. Three cohorts (INMA, KANC, RHEA) recruited during the 1st trimester of pregnancy, two through the 1st and 2nd trimesters (EDEN, MoBa), while in the BiB cohort women were recruited between weeks 26 and 28 of gestation (2nd/3rd trimesters). Inclusion and exclusion criteria varied between cohorts, as described in Table 1. All cohorts included at least one follow-up point during pregnancy, one at birth, and several after delivery.
Based on these six existing cohorts, HELIX used a multilevel study design, drawing on nested study populations for data collection of different intensities (Annex I, Figure 1): 1) the entire cohort in which factors arising primarily from outdoor exposures were assessed through geospatial models and linked to existing health outcome data; 2) a subcohort in which one new follow-up examination of the children between ages 6 and 11 years was carried out in order to assess child health outcomes and to fully characterize different areas of the exposome through questionnaires, biological sample collection, and biomarker and omics measurements; and 3) two panel studies in children and pregnant women to characterise in depth the variability in exposure biomarkers and omics biomarkers, individual exposure-related behaviours, and personal exposures.
The study population for the entire HELIX cohort includes 31,472 women who had singleton deliveries between 1999 and 2010, and for whom exposure to ambient air pollution during pregnancy had been estimated as part of the European Study of Cohorts for Air Pollution Effects (ESCAPE) project. The entire cohort includes nine regions from the six cohorts; we included only regions where geographic data were available to calculate air pollution levels and built environment indicators (Table 1). This meant, for example, that the city of Oslo and not the whole of the national MoBa cohort was included, and that only the Gipuzkoa, Sabadell and Valencia regions of the INMA study were included. In the other cohorts, women residing outside the main urban areas were excluded for the same reason.
In this study population, data on many variables had been collected in the individual cohorts during previous data collection points (during pregnancy and between birth and five years of age). Existing data included information on certain exposures (e.g.: maternal tobacco smoking during pregnancy, environmental tobacco smoke), key covariates (e.g.: pregnancy complications, maternal and child diet, maternal and child physical activity, child sleep, breastfeeding, other health related behaviours, indicators of socioeconomic status), and health and development outcomes. As part of HELIX, relevant datasets from all 31,472 mother-child pairs were transferred from the six cohorts to the central HELIX data warehouse located at the Barcelona Institute for Global Health (ISGlobal). Through data harmonisation these cohort-specific variables were converted to harmonised variables. This process involved summarising, checking, and matching the specific variable cohort-by-cohort and deciding a common coding system appropriate to each variable. Specific expert working groups throughout the HELIX consortium advised on the harmonisation rules for each variable. The child health and developmental outcomes harmonised as part of HELIX include birth outcomes, growth and obesity related outcomes, blood pressure, neurodevelopment, and respiratory health between birth and 5 years of age.
From the entire cohort, a subcohort of mother-child pairs was selected to be fully characterised for a broad suite of environmental exposures and “omics” data, to be clinically examined, and to have biological samples collected. A new follow-up visit was organised for these mother-child pairs between December 2013 and February 2016. Subcohort subjects were recruited from within the entire cohorts such that there were approximately 200 mother-child pairs from each of the 6 cohorts. Subcohort recruitment in the EDEN cohort was restricted to the Poitiers area and in the INMA cohort to the city of Sabadell.
Eligibility criteria for inclusion in the subcohort were: a) age 6-11 years at the time of the visit, with a preference for ages 7-9 years if possible; b) sufficient stored pregnancy blood and urine samples available for analysis of prenatal exposure biomarkers; c) complete address history available from first to last follow-up point; d) no serious health problems that may affect the performance of the clinical testing or impact the volunteer's safety (e.g. acute respiratory infection). In addition, the selection considered whether data on important covariates (diet, socio-economic factors) were available. Each cohort selected participants at random from the eligible pool in the entire cohort and invited them to participate in this subcohort until the required number of participants was reached. In total 1301 mother-child pairs with complete questionnaire and clinical examination data, and urine and blood samples, were included in the HELIX subcohort (Annex I, Figure 1). Several cohorts then invited and examined further subjects (N=322) following the same protocol, but these were not included in the measurement of exposure biomarkers for the HELIX study (Annex I, Figure 1).
The new follow-up visits for the subcohort took place in the six study centres at a local hospital, a primary care centre, or at the National Institute for Public Health (NIPH) in Oslo. During the follow-up examination, trained nurses interviewed the mothers, carried out health examinations of the children and collected biological samples using standardised operating procedures. Extensive cross-cultural questionnaires for use in all cohorts were developed and translated. These include questions required for accurate exposure assessment e.g. behavioral, diet and social data.
Intensive repeat panel studies collected data on short-term temporal variability in exposure biomarkers and omics biomarkers, individual behaviours (physical activity, mobility), and personal and indoor exposures. The child panel study included children from the HELIX subcohort (n=157, from all cohorts except MoBa) who lived in a first floor apartment or private house and were sampled following a maximum variation sampling strategy to high traffic-density exposure at home address. The pregnancy panel study included pregnant women from outside the cohorts in three cities, Barcelona, Grenoble, and Oslo (N=158). The inclusion criteria for these pregnant women were: to be 18 years or older at the start of pregnancy, to have a singleton pregnancy, to be living in the study area until delivery, and to have the first visit before the end of gestational week 20. Participants in the child panel study were followed for one week in two seasons, whereas in the pregnancy panel study the participants were followed for one week in two trimesters. In the child panel, the last day of the first week coincided with the subcohort examination, detailed above. Panel study subjects provided daily urine samples and at the end of each monitoring week blood samples were collected following the same procedures as for the subcohort. The panel study subjects wore smartphones for mobility and physical activity monitoring, electronic wristband UV dosimeters, active PM2.5 Cyclone samplers, and MicroAthelometers for continuous black carbon monitoring (Annex 1, Figure 2). .
Relevant datasets from all 31,472 mother-child pairs were transferred from the six cohorts to the central HELIX data warehouse located at ISGlobal. The data warehouse has been established in a format that allows future use beyond the project lifespan (2013-2017) as an accessible resource for collaborative research involving researchers external to the project. Access to HELIX data is based on approval by the HELIX Project Executive Committee and by the individual cohorts, who will evaluate potential overlap with ongoing work, adequacy of data protection plans, logistic and financial consequences, and adequacy of authorship and acknowledgement plans. We encourage interested researchers to contact us to set up collaborations. Further details on the content of the data warehouse and procedures for external access are described on the project website (http://www.projecthelix.eu/).
Project Results:
WP1 - Novel Tools for Individual Exposures
The aim of WP1 was to develop and apply novel tools and methods to obtain robust estimates of exposures to persistent and non-persistent pollutants in food, consumer products, water and indoor air, during pregnancy and early childhood. To improve the individual assessed exposures and reduce measurement error, sensitive methods for predicting exposure and repeated biomarker measurements, were applied. For all these exposures, methods that integrate exposure biomarkers and personal or environmental monitoring with knowledge on individual variability and behaviours, would lead to more accurate characterization of exposure. This required detailed characterization of short-term temporal variability and individual behaviours that influence exposure. The aim of WP1 was broken down in specific tasks;
A) To collect biological samples and exposure data in the HELIX subcohort;
B) To carry out the children and pregnancy exposome variability panel studies;
C) To determine biomarkers of exposure to persistent and non-persistent pollutants;
D) To develop and validate exposure prediction models for Disinfection By-Products (DBPs) and indoor air pollutants;
E) To assign exposure estimates to the subjects in the study and prepare a database with exposure
After years of thorough planning and field work the establishment of the HELIX subcohort as well as the panel studies were completed early in 2016. The final number of subjects who completed the fieldwork was 1623 children in the subcohort, and 157 children and 158 pregnant women in the panel studies respectively.
To reduce uncertainty and have as comparable results as possible, HELIX used one laboratory for all new measurements of the individual chemical exposure biomarkers. The samples were randomized into batches before chemical analyses. Data on biomarkers of exposure from pregnant women already existing were retrieved from the cohorts and undergone thorough quality control.
Biomarkers of contaminant exposure were measured in appropriate biological samples collected from the children at age 6-11 years and in samples previously collected from mothers during pregnancy or from the neonates during delivery (cord blood) and stored in cohort biobanks. Chemical assays were conducted in the laboratory at the Department of Environmental Exposure and Epidemiology at the NIPH, apart from analyses of metals/elements and cotinine, creatinine and blood lipids, which were subcontracted to ALS Laboratory Group Norway AS and Dr. Fürst Medisinsk Laboratorium AS, respectively. Biomarkers include: organochlorine compounds and brominated compounds, perfluoroalkyl substances and metals in blood, and non-persistent chemicals (phthalate metabolites, phenols, organophosphate pesticide metabolites, and cotinine) in urine samples (Table 2). Urine samples of the night before the visit and the first morning void on the day of the visit were combined to provide a slightly longer-term exposure assessment than can be achieved with one spot urine sample [M Casas et al - in review].
Table 3 gives and overview of the concentrations measured in samples taken from the mother during pregnancy and in the child 6-11 years later (the subcohort). This forms the basis for the individual early life chemical exposome to be included in work in the other work packages in HELIX. Figure 3 shows this “chemical exposome” for the persistent organic pollutants (the POPs exposome) measured in blood in children and mothers in the six cohorts. The concentrations of the exposure markers measured in the panel studies are presented in Annex I, Figure 4.
Concentrations of drinking water disinfection by-products (DBPs) during pregnancy were estimated from water company concentration and distribution data as part of the water contaminants and still birth, congenital anomalies, birth weight, preterm delivery (HiWate) project in four of the cohorts (BiB, KANC, INMA, RHEA). For EDEN and MoBa we followed the same methodology to obtain estimates during pregnancy. Data was not sufficiently complete to estimate child exposure to DBPs. Indoor air concentrations of nitrogen dioxide (NO2), particulate matter <2.5μm (PM2.5) particulate matter absorbance (PMabs), benzene, and toluene, ethylbenzene, xylene (TEX) were estimated by combining measurements in the homes of a subgroup of children during the two periods of the nested panel studies (see below) with questionnaire data from the subcohort.
Conclusions
This project has demonstrated that it is possible to perform a harmonized, though extensive, fieldwork using the same protocols for sample collection and clinical examination in six European mother-child cohorts. A wide range of exposure biomarkers were determined with high detection frequencies in all individuals in the entire subcohort. Thus, this study presents harmonized and completely comparable biomonitoring data for a plethora of environmental contaminants in children from several European countries, as well as comparable data for their mothers in samples taken during pregnancy. The concentrations found in the maternal samples were in general higher than concentrations measured in the children’s samples. For most of the persistent compounds the correlation between maternal and children’s levels was high, while considerably lower correlations were observed for most non-persistent compounds. The concentrations were significantly different between cohorts for more or less all compounds.
From the in depth, panel studies we have concluded that there is a substantial variability in exposure during pregnancy and over a year in school-age children for the majority of phthalate metabolites, phenols, and OP pesticide metabolites. The sample available for all HELIX subcohort children predicts well the annual exposure of most phthalates and oxybenzone, but it is not a good predictor of the other phenols and none of the OP pesticides. The quantification of the temporal variability of the non-persistent compounds can be used in Exposome studies to reduce bias. For most compounds 3-5 pools of 15/20 urines each would be necessary to obtain excellent reliability (Intraclass correlation coefficient>0.8).
Overall, HELIX has shown that children across Europe are exposed to a wide range of environmental chemicals in fetal life and childhood, and that the exposure varies between countries and by compounds. HELIX comprises a unique dataset to study exposure to chemical mixtures on an individual level and assess impact on health.
WP2 - Novel Tools for Outdoor Exposures
WP2 aimed to develop and apply novel tools and methods to obtain robust estimates of chemical and physical exposures in the outdoor environment ("the outdoor exposome"), focusing on key outdoor exposures (outdoor air pollutants, noise, green space, UV radiation).
In the entire cohort and subcohort, a GIS environment for the nine study areas was constructed, and, based on residential address histories, exposure estimates were assigned for ambient air pollutants, road traffic noise levels, surrounding (natural spaces green and blue spaces), built environment, ultraviolet (UV) radiation, and meteorological variables during pregnancy and childhood. These estimates build on existing land-use regression air pollution models (ESCAPE project), city noise maps, land use maps (“Urban Atlas” by European Environmental Protection Agency), raster maps of the normalized difference vegetation index (NDVI), raster maps of land surface temperature, building density, population density, connectivity, walkability and public bus transport map information for the built environment, and meteorological data, as described in more detail elsewhere [Robinson et al. – submitted]. Exposures where assigned within GIS techniques to all geocoded addresses of cohort subjects during the pregnancy, at birth, and at postnatal follow-up points. Data from existing regulatory monitors were used to back extrapolate ambient air pollution exposure models. The estimates for these outdoor exposures were calculated for the prenatal period and several postnatal periods up to the HELIX subcohort follow-up time point. Table 4 (Annex I) shows a summary of all GIS variables have been generated.
Daily measurements of temperature, humidity and pressure was obtained from a local weather station in each study area and averaged over pregnancy. Daily measurements of UV radiation (as erythemal UV, DNA damaging UV and vitamin D UV dose) at 0.5 x 0.5 degree resolution was obtained from the Global Ozone Monitoring Experiment onboard the ERS-2 (European Remote Sensing) satellite and averaged over pregnancy.
For assessment of air pollutants, including particulate matter with an aerodynamic diameter of less than 2.5 μm (PM2.5) and of less than 10 µm (PM10), nitrogen dioxide (NO2) and nitrogen oxides (NOX), we used land use regression (LUR) or dispersion models, temporally adjusted to measurements made in local back ground monitoring stations and averaged over the whole pregnancy period. For most cities we used site-specific LUR models developed in the context of the ESCAPE project. In Bradford, assessment for PM2.5 and PM10 was made based on the ESCAPE LUR model developed in the Thames Valley region of the UK and adjusted for background PM levels from monitoring stations in Bradford. The ESCAPE European -wide LUR model was applied for PM2.5 in Nancy, Poitiers, Gizpukoa and Valencia, corrected for local background monitoring data. In Gipuzkoa and Valencia, PM10 estimates were made based on local ratios to PM2.5 estimates. In Nancy and Poitiers, dispersion models were used to assess NO2 and PM10 exposure. In Rhea, new LUR models were developed for NO2 and NOX, incorporating new road traffic intensity variables collected from a new fieldwork campaign conducted at 80 monitoring points around the city (see supplementary information).
Noise levels Lden (day evening and night average) were derived from noise maps produced in each local municipality under the European Noise Directive. To improve comparability between centres, the values were categorized into six categories for analysis. In Heraklion, estimates on noise were newly modelled following new fieldwork (see supplementary information).
We followed the PHENOTYPE protocol to measure the surrounding vegetation, i.e trees, shrubs and parkland, and applied the Normalized Difference Vegetation Index (NDVI) derived from the Landsat 4–5 Thematic Mapper (TM) satellite images at 30m × 30m resolution (US Geology Survey, 2011). NDVI is an indicator of greenness based on land surface reflectance of visible (red) and near-infrared parts of spectrum and ranges between −1 and 1 with higher numbers indicating more greenness. To achieve maximum exposure contrast, we looked for available cloud-free Landsat TM images during the period between May and August for years relevant to our period of study and calculated greenness within 100, 300 and 500 meter buffers around each address. We calculated access to major green spaces (parks or countryside) and blue spaces (bodies of water) as the straight line distance from the home to nearest blue or green space with an area greater than 5000 m2 from topographical maps (urban atlas 2006 or local sources).
Topological maps for the following built environment indicators were obtained from local authorities or from Europe wide sources. Traffic density indicators (traffic density on nearest street, traffic load on major road within 100m and inverse distance to nearest major road) were calculated from road network maps following the ESCAPE protocol. Building density was calculated within 100 and 300m buffers by dividing the area of building cover (m2) by the area of buffer (km2). Population density was calculated as the number of inhabitants per km2 surrounding the home address. Street connectivity was calculated as the number of intersections inside 100m and 300m buffers, divided by the area (km2) of each buffer. Facility richness index was calculated as the number of different facility types present divided by the maximum potential number of facility types specified, in a buffer of 300 meters, giving a score of 0 to 1. Land use Shannon's Evenness Index (SEI) was calculated as the proportional abundance of each land use type multiplied by that proportion, divided by the logarithm of the number of land use types, in a buffer of 300 meters, giving a score of 0 to 1. We developed an indicator of walkability, adapted from the previous walkability indexes, calculated as the mean of the deciles of population density, street connectivity, facility richness index and land use SEI within 300m buffers, giving a walkability score ranging from 0 to 1.
Estimates were generated during pregnancy and follow up of the subjects in the main cohort, for all the children of the subcohort and for the panel studies of the children and pregnant women.
Results from estimates made during pregnancy showed that mean noise levels ranged from 49.6 dB (Kaunas) to 63.9 dB (Heraklion), mean NO2 levels ranged from 13.6 ug/m3 (Heraklion) to 43.2 ug/m3 (Sabadell), mean walkability score ranged from 0.22 (Kaunas) to 0.32 (Valencia) and mean Normalized Difference Vegetation Index (NDVI, 300m buffer) ranged from 0.21 in Heraklion to 0.51 in Oslo. NDVI was correlated with NO2 (-0.42) noise (-0.26) building density (-0.74) and UV radiation (-0.13) among other indicators. Four PCs explained more than half of variation in the urban exposome. We observed considerable heterogeneity in social patterning of the urban exposome across cities. For example, high SES women lived in greener, less noisy and polluted areas in Bradford (0.18 standard deviations of PC1, 95% CI: 0.15 0.22) but the reverse was observed in Oslo (−0.25 standard deviations of PC1, 95% CI: −0.32 −0.18).
Furthermore, participants of panel studies were monitored twice regarding their mobility, physical activity, and personal exposure to air pollution, noise, ultraviolet (UV) light, and natural outdoor environment, using a newly developed personal exposure monitoring (PEM) kit. The two repeated measurements were obtained during two separate normal weeks (i.e. school and working weeks) with 6 months of difference. The PEM kit was composed of: (i) a belt with an attached smartphone and accelerometer; (ii) a wrist UV dosimeter; (iii) a small backpack fitted with one gravimetric sampler to measure particulate matter with aerodynamic diameter of 2.5 µm or less (PM2.5); and one real-time sampler to measure black carbon (BC); both with the inlets attached to one handle of the backpack at the breathing zone; and (iv) two extra gravimetric samplers for PM2.5: one placed at the living room and the other outside of the main window or balcony of the house.
Geographical location was obtained from the ExpoApp application, which was running in a smartphone GT-S5360. Physical activity was assessed with a wGT3X-BT tri-axial accelerometer (ActiGraph, LLC, USA) and the ExpoApp. The wGT3X-BT and ExpoApp were set to sample at 30 and 10 Hz, respectively. The physical activity information obtained was the wearing time, intensity, duration, and frequency of the physical activity at minute resolution (Choi et al. 2012; Crouter et al. 2010, 2013; Donaire-Gonzalez et al. 2013). The intensity of physical activity together with the age, sex and weight of individuals was used to estimate the inhaled rate per minute, using the existing equations from the Environmental Protection Agency
We found that there is considerable variation in the personal urban early life exposome with a considerable percentage exposed to high levels. More than 10% of pregnant women and children were exposed to fine particles levels ≥ 25µg/m3 and ≥ 50% of them did not have any contact with NOE during one week. Moreover, children exposure levels overall were higher than those in pregnant women, particularly for UV-B doses and physical inactivity. Furthermore, we found that most of the personal exposures were not correlated but the relationships were different in children than in pregnant women. Finally most of the variability of the exposures was explained by city characteristics, especially in children.
Conclusions
We estimated and measured successfully a large range of exposures including air pollution, UV, green space, built environment, meteorology, and noise for the participants at different time points. The results show that there is considerable variation in the exposure levels between subjects and cities depending on the exposure of interest. The estimates will be used in the epidemiological analyses.
WP3 - Integrating Multiple Exposures
WP3 aimed to integrate data on multiple exposures, exposure determinants, and exposure variability (temporal, individual, toxicokinetic), in order to define multiple exposure patterns and describe exposure uncertainties.
Predicting target tissue dose through PBPK modelling
The objective of this task was to analyze the measured biomarker concentrations (from HELIX subcohort and panel studies) using physiologically based pharmacokinetic (PBPK) models to simulate the dosimetry in the target tissues. Our methodology was applied to compounds having different half-lives: PFOS and PFOA for semi-persistent compounds applied to the subcohort of 1,200 mother-child pairs, and a phthalate (DEHP, di(2-ethylhexyl) phthalate) for non-persistent compounds applied to the panel study of pregnant women (about 150 women).
PFOS: We chose to adapt a generic and lifetime PBPK model to describe the toxicokinetics of PFOS (Beaudouin, et al. 2010). Several changes were made to fit the model to the characteristics of PFAS and to improve the physiological description. The exposures of the mother and child were reconstructed using the measured biomarkers and the PBPK model. A high diversity of toxicokinetic profiles has been estimated based on the individual information collected through the questionnaires. The main determinants of the child exposure are: the levels at birth (correlated with the mother’s biomarker), the breastfeeding that can result in high levels in child plasma, and the measured biomarker itself. Using the reconstructed exposure scenarios for the children as input, PBPK models were used to simulate target tissue dosimetry at critical time periods. Our results show that the in utero indicators are all highly correlated with the measured maternal plasma concentration at the time of pregnancy, and that actual measurements during pregnancy and at the age of 6 to 9 do not correlate well with the predicted internal exposure during the first years of life (birth to 4 years).
DEHP: A toxicokinetic model (Lorber et al., 2010) that simulates serum and urine concentrations of DEHP and five metabolites (MEHP, MECPP, MEHHP, MEOHP, 2cx-MMHP) using a first order dissipation equation along with routing assumptions from parents to metabolites and to the bladder for excretion via urination was used. This model was used in a reverse dosimetry approach for estimating the distribution of exposure levels in the environment that could give rise to measured biomarker concentrations in a population. In our study, estimation of DEHP exposure by reverse dosimetry was performed for each pregnant woman. We estimated by reverse dosimetry the exposure to DEHP (in µg/kg/d) that may lead to the concentrations of DEHP metabolites measured in our study for three different scenarios of exposure. We obtained distributions of estimated exposure for each of the 3 following scenarios. The reverse modelling showed that, the mean daily intakes (DI) estimated for the women from Barcelona and from Oslo are quite similar for all the scenarios. The diet exposure scenario shows a higher variability than the two other scenarios. The diet exposure scenario seems to better depicts the high variability of the DEHP metabolites urinary concentrations. The MEHP blood concentration predicted for the continuous exposure is constant over time whereas the concentration really variable for the two other scenarios (day exposure and diet exposure). The maximal concentration (Cmax) of MEHP serum concentration predicted is higher for the diet exposure than for the continuous. There is a factor of 4 between Cmax predicted for diet and for continuous scenarios. This major difference between scenario predictions is not observed for the Area Under the curve (AUC) of MEHP serum concentration. In fact, the values of the AUC estimated are relatively similar whatever the scenario considered. The diet exposure seems to better depict the variability of the data. To conclude, toxicokinetic model and questionnaire with diet exposure data and urination times are crucial elements for the DEHP exposure assessment and for internal DEHP metabolite prediction.
Analysis of variability and uncertainties in exposure assessments
Non-persistent exposures: To estimate the uncertainty in assessments of exposure to non-persistent pollutants and second-tobacco smoke exposure we used data from the children and pregnancy panel studies. The specific objectives were: a) to evaluate between-trimester variability in pregnant women and within-week and between-season variability in school-age children of urinary concentrations of phthalate metabolites, phenols, organophosphate (OP) pesticide metabolites, and cotinine; b) to estimate whether the last pool of 2 urines of the first follow-up week (measure available in the subcohort children) was a good measure of the mean annual concentration in children; c) to calculate the number of biospecimens needed to obtain excellent reliability (defined as an ICC of 0.80 or more) of each chemical.
In children, we determined non-persistent pollutants in the urine for 4 consecutive days’ pools of week 1 (night + first morning), in the 2 weekly pools (of all samples), and in 1 pool of night and first morning voids of the last day of week 2. In pregnant women we determined non-persistent pollutants were determined in the 2 weekly pools, and phthalates were also determined in all the first morning and night voids of the first week in 45 of these 157 women (15 from each city).
Using these samples we observed that:
a) The HELIX subcohort measurement, which corresponds to the pool of night and first morning voids of the last day of week 1, is a good predictor of exposure in the week before of the majority of phthalates and phenols, DMTP and DEP pesticides, and cotinine. Subcohort measurement however, is not a good predictor of exposure in the week before of BUPA and of the majority of OPs. The subcohort measurement (week 1) is a good predictor of exposure in the second week (approximately 6 months of difference) of the majority of phthalates, BUPA, benzophenone-3 (OXBE), and cotinine but it is not a very good predictor of the majority of phenols, particularly BPA, and OPs.
b) Phthalates (except oxoMiNP), benzophenone-3, and cotinine have a low seasonal variability in children whereas the other phenols and OPs concentrations vary a lot between seasons. In pregnant women we generally observed a high variation of non-persistent pollutants concentrations between the 2nd and 3rd trimesters of pregnancy.
c) Phthalates (except oxoMiNP) and phenols (except BUPA) have low between-day variability whereas OPs have a high between-day variability in children. In pregnany women MEP, MiBP, MnBP, and MBzP have low between-day variability whereas all the other phthalates have high between-day variability.
For most non-persistent chemicals, three daily pools of 2 urines each would be needed to obtain excellent reliability for a weekly exposure window. Four weekly pools of 15-20 urines each would be necessary to obtain excellent reliability for a yearly or pregnancy exposure windows.
The ICCs calculated for each pollutant can be integrated in the exposure-response models by using the regression-calibration method. This regression allows correcting for bias from measurement error in generalized linear models.
Outdoor exposures: To estimate the uncertainty of outdoor exposures we also used data from the children and pregnancy panel studies. Specific objective was to characterize the intra- and inter-participant and intra- and inter-city variability on the personal exposure to air pollution, noise, ultraviolet (UV) light, and built/natural environments in childhood and during pregnancy. From five out of six cohorts, 150 children aged 6-9 were sampled following a maximum variation sampling strategy to high traffic-density exposure at home address. From three of the cohort regions, 150 volunteer women were recruited during her first trimester of pregnancy. With 6 months of difference, participants were continuously monitored for a normal week regarding its mobility, physical activity, and personal exposure to air pollution, noise, ultraviolet light, and built/natural environment, using a personal exposure monitoring (PEM) kit. Our findings reflect that:
a. Exposure to air pollution of children and pregnant women is different because children are more exposed to particulate matter, while pregnant women are more exposed to black carbon.
b. Exposure to air pollution is determined mainly by the activity patterns of participants’ daily life rather than by individual or city characteristics.
c. The need of epidemiological studies to include the pattern of daily life activity of participants into their air pollution exposure to mitigate the exposure misclassification.
d. The need for research to include personal exposure assessment and the importance of having access to accurate, inexpensive and not burdensome personal exposure tools.
Factor analysis of multiple exposure patterns
The objective of this task was to define exposure patterns in the multiple exposure data collected in WP 1 and WP2, using a factor analysis approach that will create a reduced set of continuous exposure indices, each of them composed of exposures that tend to occur simultaneously in the population.
The outdoor exposome collected in the HELIX project was composed of 26 different exposures belonging to 7 families of exposures (noise, air pollution, traffic, meteorological variables, built environment, green space and blue space), and was available for 27,921 pregnant women form 9 cities (Bradford, United Kingdom; Poitiers and Nancy, France; Sabadell, Valencia and Gizpukoa, Spain; Kaunas, Lithuania; Oslo, Norway; Heraklion, Greece). The full exposome was composed of 208 exposures belonging to 15 families of exposures, and was available for 1285 participants. Exposure variables were transformed when needed to have a symmetric distribution. To maximize the use of available information and provide unbiased estimates, we conducted multiple imputation of missing values using the method of chained equations. First, we displayed correlation heat maps for all exposures, overall and by cohort. Subsequently, we standardized exposures by dividing them by their respective standard deviation. Those transformed variables were used to conduct a principal component analysis. We calculated the percent of explained variance as a function of the number of principal components and displayed it graphically. The number of principal components to be retained was selected upon exploration of the latter graphical display. Varimax rotation was applied to the final solution. The loadings of each principal component were examined in order to label each principal component. The results of these analyses highlighted the following points:
a) Although correlations within the same exposure family can be high, correlations between exposures from different families were low. This supports that epidemiological studies focusing on a reduced set of exposures may not be confounded by having omitted exposures from other families.
b) The exposome is high-dimensional, as it cannot be reduced to a small set of principal components. Even the outdoor exposome required 15 components to explain most of the variability. In the full exposome, up to 60 components were needed to explain most of the variation.
c) Between-city differences influence the correlations. Correlations between exposures tend to be larger when city is not controlled for, as they also reflect between-city correlations. The first principal components reflect the differences between cities.
Model-based clustering approach to identify groups of subjects with similar pattern of exposure
The aim this task was to identify clusters of participants sharing a similar exposome. Continuous variables were transformed to achieve symmetric distributions. Since we were not interested in capturing between-cohort correlations (e.g. southern cohorts having warmer temperatures and higher air pollution), we subtracted cohort means from each exposure prior to the analyses. Continuous variables were standardized by dividing by the global standard deviation. The dataset contained continuous and binary variables, so clustering techniques that allow for mixed data were used. Several clustering techniques were applied to the data, as there is no single method that is guaranteed to outperform the other methods in all instances. In particular, we used Bayesian model based clustering, as implemented in the R statistical package PReMiuM1; partitioning around mediods (PAM) based on Gower distances, which accommodate continuous and binary variables (https://www.r-bloggers.com/clustering-mixed-data-types-in-r/); and hierarchical clustering based on Gower distance, as implemented in the R function hclust(). PReMiuM finds the optimal number of clusters using the Silhouette width, and it also performs variable selection. For the PAM method, we also calculated the Silhouette width. For the hierarchical clustering, we used the bootstrap method implemented in the pvclust R package.
In summary, the three clustering methods used did not provide useful solutions for the ultimate aim of the HELIX project, which was to link exposure clusters to health outcomes. Two-cluster solutions do not provide enough variability to capture different exposure patterns. Besides, they were highly collinear with cohort, and our analyses on the effects on health will adjust for cohort. The solutions of model based clustering, which provided a large number of clusters, were not useful either, as the clusters had too small sample sizes, and the solutions were again highly collinear with cohort. The conclusion from this part of the analysis is that clustering of exposures may not be adequate for the following analyses on the effects of the exposome on health outcomes. Other techniques, such as the principal component analysis conducted in the previous task may provide better alternatives.
Conclusions
WP3 focused on the integration of data on multiple exposures, exposure determinants, and exposure variability (temporal, individual, toxicokinetic), in order to define multiple exposure patterns and describe exposure uncertainties. Main conclusions of this WP are:
1) The analysis of the biomarkers and questionnaire data with PBPK models allows reconstructing the external exposure of semi-persistent compounds (PFAS) in children and simulating target tissue dosimetry at critical time periods during early life. Our results showed that an adequate use of individual information (e.g. breastfeeding) with a PBPK model can rebuild realistic exposure scenarios. Accounting for this inter-individual variability can reduce uncertainty and may increase the power of association analyses between adverse effects and exposure. For non-persistent compounds (DEHP), we highlighted the importance of individualizing the exposure scenario to adequately estimate the maximum exposure of the individuals at certain times of the day.
2) In terms of the characterisation of uncertainties and variabilities, for most non-persistent chemicals, multiple pools of multiple urines would be needed to obtain excellent reliability in exposure assessment; for example, 4 pools of 15-20 urines each would be needed to get a reliable estimate over the entire pregnancy. The variability in environmental outdoor exposures was mainly determined by the activity patterns of participants’ daily life rather than by individual or city characteristics. These results confirm the need of using personal exposure assessment methods for outdoor exposures in epidemiological studies and the importance of having access to accurate, inexpensive and not burdensome personal exposure tools.
3) In terms of exposure patterns: The exposome is high-dimensional, as it cannot be reduced to a small set of principal components or clusters; In fact, clustering of exposures may not be adequate for multicentre analyses on the effects of the exposome on health outcomes as location is a strong determinant of one’s personal exposome.
4) Although correlations within the same exposure family can be high, correlations between exposures from different families were low. This supports that epidemiological studies focusing on a reduced set of exposures may not be confounded by having omitted exposures from other families.
Future Exposome studies should continue refining exposure assessment through repeated collection of biospecimens and personal dosimeters in even larger populations.
WP4 - Integrating Molecular Exposure Signatures
The objective of WP4 was to determine molecular signatures associated with environmental exposures through analysis of profiles of metabolites, proteins, RNA transcripts, and DNA methylation. Specific objectives were:
▪ To generate molecular profiles in biological samples from the cohort studies using optimised protocols
▪ To correlate specific exposures and exposure clusters to molecular profile data
▪ To integrate information from molecular profiles using pathway analysis approaches to define biological pathways associated with exposure
Omics data generation and quality control
Urinary metabolic profiles were analysed on a 14.1 Tesla (600MHz 1H) nuclear magnetic resonance spectrometer at Imperial College London (ICL). Most samples analysed (1273/1366) were a pool of samples taken from each child in the morning and the night before. The targeted metabolomics AbsoluteIDQTM p180 Kit (BIOCRATES Life Sciences AG) was used to profile the serum samples from the sub-cohort (n=1364 samples). The kit allows the targeted analysis of 188 metabolites in the metabolite classes of amino acids, biogenic amines, acylcarnitines, glycerophospholipids, sphingolipids and sum of hexoses.
Buffy coat DNA was extracted using the Chemagen kit (Perkin Elmer) in batches of 12 samples. Samples were extracted by cohort. DNA concentration was determined in a NanoDrop 1000 UV-Vis Spectrophotometer (ThermoScientific) and with Quant-iT™ PicoGreen® dsDNA Assay Kit (Life Technologies). 700 ng of DNA were bisulfite-converted using the EZ 96-DNA methylation kit following the manufacturer’s standard protocol, and DNA methylation was assessed with the Infinium HumanMethylation450 beadchip from Illumina, following manufacturer’s protocol. DNA methylation data was pre-processed using the minfi package . After sample and probe quality control ), data was normalized with the functional normalization method, which also includes Noob background subtraction and dye-bias correction . The final dataset consisted of 1,347 HELIX samples representing 1,192 subjects and 485,512 probes (or 386,518 probes after filtering of probes with SNPs, probes that cross-hybridize and probes in sexual chromosomes).
RNA was extracted from 1,382 HELIX samples (and 308 extra HELIX samples) using the MagMAX for Stabilized Blood Tubes RNA Isolation Kit (ThermoFisher). Mean values for the RIN, concentration (ng/ul) and Nanodrop 260/230 ratio were: 7.05 109.07 and 2.15. Gene expression, including coding and non-coding transcripts, was assessed with the Affymetrix Human Transcriptome Array 2.0 ST arrays (HTA 2.0). Amplified and biotynylated sense-strand DNA targets were generated from total RNA of the 1,304 samples with good RNA quality and hybridized on Affymetrix HTA 2.0 arrays. Data was normalized with the GCCN (SST-RMA) algorithm at the gene and transcript level. Annotation to transcripts clusters was done with the ExpressionConsole software using the HTA-2_0 Transcript Cluster Annotations Release na36. Control probes and probes in sexual chromosomes or probes without chromosome information were excluded. Probes with a call rate <70%, based on DABG (Detected Above Background) p value < 0.05 were excluded from the analysis.
Expression miRNA levels of 1,087 samples with good RNA quality were analysed using the SurePrint Human miRNA Microarray rel. 21 (Agilent) following Agilent's recommendations. Briefly, RNA samples were concentrated or evaporated in order to reach the required concentration using a vacuum equipment (SpeedVac). The miRNA Complete Labeling and Hyb kit generates fluorescently-labeled miRNA with a sample input of 100 ng of total RNA. This method involves the ligation of one Cyanine 3-pCp molecule to the 3' end of a RNA molecule. Agilent SurePrint G3 Human miRNA microarrays were hybridized following the Agilent Microarray Hybridization Chamber User Guide. After an initial quality control based on laboratory parameters, miRNA levels were normalized using the least variant set (LVS) method with background correction using the Normexp method in limma package . After normalization, miRNAs with a call rate <70%, based on Agilent’s p value, and miRNAs in sexual chromosomes were filtered out. The final dataset consisted of 1,078 samples and 330 miRNAs.
A set of 43 proteins were selected a priori based on the literature and on the Luminex kits commercially available from Life Technologies and Millipore. Three kits were selected, which assessed a total 50 measurements: Cytokines 30-plex (Cat #. LHC6003M), Apoliprotein 5-plex (LHP0001M) and Adipokine 15-plex (LHC0017M). For protein quantification, an 8-point calibration curve per plate was performed with protein standards provided in the Luminex kit and following the procedures described in the standard procedures described by the vendor. The % of coefficients of variation (% CV) for each protein estimated by plate and then averaged ranged from 3.42% to 36%. Seven proteins were removed because they had <30% of measurements in the linear range. For the 36 proteins that passed the QC, data was log transformed to reach normal distribution. Then, the plate batch effect was corrected by subtracting for each individual and each protein the difference between the overall protein average minus the plate specific protein average. Finally, values >LOQ2 (upper limit of quantification) were set to NA and values
For further details including procedures for data quality control please see the previous deliverable report (see Deliverable Report D4.4 “Completed omics database, selection of targets for health analysis”).
ExWAS analyses
Linear regression models were generated for each exposure-omics biomarker pair using omicRexposome, which was developed as part of HELIX project (https://github.com/isglobal-brge/omicRexposome). Models were adjusted for age, sex and cohort. Surrogate variable analysis (SVA) was applied to correct for main unwanted variability derived from technical batch, blood cell type proportions or others. This approach was applied only on the following datasets: miRNA, gene expression and methylation. The significance threshold was adjusted for multiple testing using Bonferroni correction considering either the number of omics markers for each platform (‘O’) or the number of omics-exposure biomarker pairs (‘OE’).
Omics data: number of samples and features
The final number of samples and features for the HELIX subcohort –omics can be seen in Table 5,1 (Annex I). The overlap between samples with several omics data can be seen in Table 5.2 (Annex I); in total, 874 samples have available all omics data planned in HELIX in addition to the exposome and health data.
ExWAS: metabolomics, proteomics and miRNAs
Our initial ExWAS analyses focused on integration of the exposome with proteomics, metabolomics and miRNA profiles as these were available in advance of other omics and with less features require less computational expense to complete. ExWAS analysis revealed a number of significant associations between molecular features and exposures (Figure 5, Annex I), even after adjustment for multiple testing, considering different thresholds. The highest number of significant associations was found with the postnatal exposome, especially with the serum metabolome. Few associations (N=4) overlapped between pregnancy and postnatal exposome.
Figures 6-9 (Annex I) illustrate examples of the emergent exposure/biomarker signatures revealed by ExWAS analysis. Serum metabolites exhibited strong positive correlations to levels of PFASs as well as levels of arsenic (As) and mercury (Hg) (Figure 6). Examination of the associated metabolites indicated an over-representation of phosphatidylcholine lipids containing polyunsaturated fatty acids (nomenclature PC total number of carbons in fatty acids: total number of double bonds in fatty acids). Since humans cannot desaturate fatty acids beyond carbon 9, a key supply of these essential fatty acids is the diet. Polyunsaturated fatty acids levels are high in certain plant products and also fish and other seafood. The latter can accumulate toxins such as heavy metals and therefore the observed signature may be driven in part by common routes of exposure. A similar pattern was observed for urinary metabolome (Figure 7). Arsenic and mercury were associated with trimethylamine-N-oxide, a metabolite found in fish; while organophospate pesticides were associated with hippurate and proline betaine, metabolites incorporated from fruits and vegetables.
We also observed a cluster of inverse correlations between exposure to organochlorines (OCs), and to a lesser degree PBDEs, with circulating levels of the appetite-regulating hormone leptin, and two important pro-inflammatory, adiposity-associated cytokines (adipokines), IL-6 & IL-1beta (Figure 8). Associations were attenuated after adjustment for child body mass index, which suggests that fat mass acts as a confounder. Blood concentration of lipophillic persistent organic pollutants such as OCs and PBDEs is dependent on total body fat (ie. given the same exposure, obese children have lower blood concentration of OCs), and on the other hand fat mass is the main producer of adipokines.
Blood miRNAs were associated with meteorological exposures such as temperature and ultraviolet radiation (Figure 9). Further investigation is needed to discard potential cohort effects.
NO2, mercury and HCB: methylome and transcriptome
We are working on the analysis of the full exposome (ExWAS) and the methylome and transcriptome. At this point, we have finalized the analysis of three “reference” exposures: NO2, mercury and HCB. Postnatal NO2 levels were associated with blood DNA methylation levels at 44 CpG sites. These, were enriched in genes related to blood pressure and body mass index regulation. In contrast prenatal NO2 levels did not give many associations at age 7-9 years (n=4), and they did not overlap with associations observed for postnatal NO2. Exposure to mercury was not associated with methylation, while prenatal and postnatal HCB was associated with very few CpG sites (n=4-5). We did not find any statistically significant association between these three exposures and the transcriptome.
SUMMARY
In the HELIX project we have successfully generated complete molecular profile data sets comprising of urinary and serum metabolomics, plasma proteomics, blood cell DNA methylation, transcriptomics and miRNA data for 874 children in the HELIX subcohort, with up to 1198 children’s samples analysed for individual omics platforms. An EXposure Wide Association Study (ExWAS) was conducted with metabolomics, proteomics and miRNA data and the child-matched full exposome dataset from HELIX. These analyses identified several significant clusters of associations, including: serum polyunsaturated glycerophospholipids with exposure to heavy metals and PFASs; urinary trimethylamine-N-oxide with exposure to heavy metals and urinary hippurate and proline betaine with organophospate pesticides. These signatures are likely to represent common routes of exposure, e.g. fish and seafood as a common source of polyunsaturated fatty acids and metals. We also found associations between OCs and PBDEs with plasma adipokines, which can reflect a link with fat mass, e.g. fat mass as a driver of adipokine expression and storage of lipophilic chemicals. A similar statistical analysis followed with functional enrichment is ongoing for the methylome and transcriptome and the full exposome.
WP5 - Linking the Exposome to Child Health
WP5 aimed to characterize the effect of the Exposome on specific highly prevalent child health outcomes. These health outcomes are pre- and postnatal growth and obesity, asthma and respiratory function, and neurodevelopment. Objectives of WP5 were to carry out the health outcome examinations in the HELIX subcohort; to design and maintain the HELIX data warehouse; to perform a simulation study to compare the efficiency of various statistical approaches considered to assess the impact of the Exposome on human health; and to provide exposure-response estimates for the association between the Exposome and child health.
Health outcome examinations were carried out using standarised protocols in the six cohorts (children aged 6-11 years). Results show large differences between cohorts in the health outcomes: For example, food allergy questionnaires showed that overall 21% of children were reported to have at least one food allergy (ever experienced), ranging from 15.6% in RHEA to 35% in INMA. The percentage of children who had ever had asthma was low in INMA (3.6%) and high in BiB (18.5%) and EDEN (20.2%). Overall, 18.8% of children were overweight and 9.9% were obese (total 27.7%). The percentage of overweight and obese children (using the age-and-sex-standardised z-scores) was highest in RHEA (37.2%) and INMA (42.3%) and lowest in MoBa (15.8%). ADHD symptoms assessed through the Conner’s rating scale were classified using the cut-off score of the 80th percentile. Using this classification, 10.1% of children in the subcohort were classified as having ADHD symptoms, ranging from 4.4% in MoBa to 15.2% in KANC. The total problems score of the CBCL, which consists of the sum of ratings on all 120 behavioural and emotional items of the CBCL, also showed that mothers in MoBa reported the lowest total score (median score 9) and mothers in KANC the highest (median score 27).
The HELIX data warehouse was constructed as a relational database created in MySQL. New data, collected through the common protocols during the subcohort and panel study fieldwork, were entered directly into an electronic database and then uploaded into the data warehouse. Questionnaires were computer-based with a direct entry to the database. All data were locally and centrally checked by examination of the ranges, distributions, means, standard deviations, outliers and logical checks. Data outliers and missing values were checked with the local cohort field workers and, where possible and relevant, replaced by correct values. All new measurements of exposure biomarkers and omics from the labs, and all exposure variables estimated through geospatial models and other methods, were added to the data warehouse as they became available.
Exposome-health associations
From a methodological perspective, most previous studies relating the Exposome to health relied on the Environment-Wide Association Study (EWAS) 5, possibly followed by a multiple regression step 6. Several other regression-based methods exist and allow accounting for a potential joint action of multiple exposures on health. Sparse Partial Least Square (sPLS) for instance has recently been used in a study of male fecundity 4, while Elastic Net (ENET) was used to link multiple environmental contaminants to birth weight1. The statistical performances of these various models in an Exposome context remain to be systematically assessed. Two- or three-way interactions between environmental exposures have been described in the literature
and statistical methods to uncover interactions among a large set of exposures have been suggested 7,8. Again, their efficiency and limitations have not been systematically assessed.
Our aims were twofold: 1) to identify agnostic statistical methods most suited for the study of the effects of the Exposome on health, both in the absence or presence of interactions between the components of the Exposome; 2) to characterize the association between the early-life Exposome and child health (birth weight, body mass index, BMI, in childhood, respiratory health, neurodevelopment, blood pressure).
Methods
Our approaches combined simulation studies based on a realistic Exposome data structure and real analyses of the unique HELIX subcohort population, in which over 100 exposure factors assessed from biomarkers (WP1) and environmental models (WP2) and belonging to over 15 exposure families have been characterized during pregnancy and in childhood.
Real analyses were performed using the statistical models that showed the best performances in simulation studies, as well as univariate descriptive approaches (EWAS). All models were adjusted for a predefined set of potential confounding factors. All continuous variables were normalized and standardized by the IQR before being analysed. Effort was made to reduce the initial set of exposures and avoid correlation >90%; when exposures were assessed at different windows (e.g. time points or buffers), we a priori selected one of them in the primary analysis. Independent analyses were performed for the exposure variables measured during prenatal and postnatal periods. All analyses were performed using the R software (www.r-project.org). We used the Rexposome package for drawing plots, mice for multiple imputation and DSA for the DSA algorithm.
5.1 Simulation studies: which approaches should be used to characterize Exposome-health relations?
Regarding the simulation studies , in the absence of interaction, the elastic net, sparse partial least-squares regression, Graphical Unit Evolutionary Stochastic Search (GUESS) Bayesian variable selection model, and the deletion/substitution/addition (DSA) algorithm showed on average over all simulation settings a sensitivity of 78% and a FDP of 37%, with minor differences between methods. The environment-wide association study (EWAS) underperformed these methods in terms of FDP (68% on average) and bias, with a higher sensitivity. When we assumed that some exposure factors acted in synergy, the models GLINTERNET and DSA had better overall performance than the other methods, with GLINTERNET having better properties in terms of selecting the true predictors (sensitivity) and of predictive ability, while DSA had a lower number of false positives. In terms of ability to capture interaction terms, GLINTERNET and DSA had again the best performances, with the same trade-off between sensitivity and false discovery proportion 9.
5.2 Which components of the Exposome were related to the health of HELIX children?
Regarding the analyses of HELIX subcohort data, EWAS and DSA models identified several associations with the health outcomes considered, several of which were consistent with the existing literature.
Birth weight: In the case of the birth weight analysis, lead was the only exposure associated with birth weight in the DSA model corrected for exposure misclassification (an effect restricted to boys, corresponding to a decrease in mean birth weight with lead exposure, Figure 14, Annex I). The only interaction term identified was between a marker of socio-economic status and blue space exposure. No exposure passed the significance threshold corrected for multiple testing of EWAS (Figure 13, Annex I); the exposures most strongly associated with birth weight were dimethyl thiophosphate (DMTP, positive association) and lead and fine particulate matter (negative associations for both exposures).
BMI: Results of the analyses linking the prenatal and postnatal exposome to BMI are presented in Table 8 and Table 9. In the prenatal exposome, exposures that showed a statistically significant association with BMI in the univariate models were active smoking (increased BMI), passive smoking (increased BMI), cotinine (increased BMI), cadmium (increased BMI), and facilities density (decreased BMI); none of these associations passed an FDR corrected p-value of 0.05. The DSA model identified prenatal exposure to cotinine, active and passive maternal smoking and facilities density as contributors to the model but out of these, only active and passive smoking showed statistically significantly associations (p<0.05) with BMI. The p-value for facility density in the DSA model was 0.07. The other indicators of adiposity (waist circumference for abdominal fat, skinfolds for subcutaneous fat, and proportion fat mass from bio-impedence) showed similar results with some small differences in the ranking of exposure variables (not shown). In the postnatal exposome analysis, 34 exposure variables showed significant associations with child BMI in the univariate model and 24 of these had a FDR corrected p-value below 0.05 (table 9, figure 16). These exposures included many POPs: all organochlorines, PBDE153 and some PFAS levels were negatively associated with BMI of the children. Indoor and outdoor air pollution exposures and passive smoking indicators were positively associated with child BMI. Copper and caesium were positively associated with BMI and cobalt and molybdenum negatively. Some dietary intakes (bakery products, fat consumption, breakfast cereals) were negatively associated with BMI. OXBE levels were positively associated with child BMI. Sleep duration was negatively associated with BMI. The DSA model selected indoor PM, copper, HCB, PBDE153 as being significantly associated with BMI of the children.
Spirometry: FEV1% was available for 1,033 (79.4%, ≥70.6% by cohort) children. Values ranged from 60.9 to 139.2 with an average (sd) value of 98.8 (13.2) and large between-cohort variations. In the restricted exposome analysis of prenatal and postnatal exposures, no exposure variable was selected when DSA was applied except for living with cat during childhood; EWAS did not identify any statistically significant exposure-outcome association when correcting for multiple testing. Without correcting for multiple testing in EWAS, 3 prenatal exposure variables were statistically significant at a 5% level: PFOA (decrease), inverse distance to nearest road (increase) and Perfluorononanoate (PFNA, decrease). When adjusting for potential confounding due to co-exposition (i.e. the 9 most significant exposure variables were jointly included in a multivariate linear model), results for the inverse distance to nearest road little varied, while for PFAS and PFNA which were highly correlated (at a 0.61 level), the estimated coefficients lowered and were not statistically significant at a 5% level (Table 10, Figure 18, Annex I). Also, not correcting for multiple testing in EWAS, 6 postnatal exposure variables were statistically significant at a 5% level: MEOHP (decrease), house crowding (decrease), Number of bus public transport mode stops in a 300m buffer around school (decrease), ethyl paraben (decrease), copper (decrease) and Mono-4-methyl-7-oxooctyl phthalate (OXOMINP, decrease). When adjusting for potential confounding due to co-exposition (i.e. the 28 most significant exposure variables were jointly included in a multivariate linear model), coefficients did not vary much, but only ETPA and house crowding remained significant (Table 9, Figure 19, Annex I).
Allergy-related outcomes: The prevalences of the different outcomes were 24, 25, 21 and 10% for itchy rash, rhinitis, eczema and food allergy, respectively. Results of the analyses linking the prenatal exposome to the allergy related outcomes itchy rash last 12 months, eczema ever, rhinitis last 12 months and food allergy ever are presented in Table 11. The EWAS model identified traffic density on nearest road during pregnancy to be positively associated with itchy rash, whereas maternal cotinine levels during pregnancy were inversely associated with itchy rash. However, none of these associations remained statistical significant after the p-value correction. With regard to
rhinitis, PM absorption levels during pregnancy were inversely associated, while inverse distance to nearest road during pregnancy and maternal mono-4-methyl-7-oxooctyl phthalate (OXOMiNP) levels were positively associated with rhinitis. None of the associations remained statistical significant after the p-value correction. None of the exposure variables in the prenatal exposome were significantly associated with eczema. Traffic density on nearest road and the day–evening–night noise level (lden) were inversely associated with food allergy. The inverse distance to nearest road during pregnancy was positively associated with food allergy. None of the associations remained statistical significant after the p-value correction. None of the prenatal exposure remained significantly associated with the allergy-related outcomes (itchy rash last 12 months, eczema ever, rhinitis last 12 months and food allergy ever). However, our findings might indicate a role of exposure in the general external environment such as air pollution, traffic and noise.
Asthma: For childhood asthma, analyses were performed on around 1300 children from whom questionnaire data was available. The prevalence of parent reported ever asthma at age 5 ranged from 3.7% (SAB cohort) to 20.2% (EDEN cohort). Results of the EWAS analyses linking the prenatal exposome to asthma are presented in Figure 20. None of the exposure variables in the prenatal exposome were significantly associated with asthma.
Neurodevelopment: Several components of neurodevelopment have been assessed in HELIX children: cognition (assessed from Raven test), internalizing and externalizing disorders, both assessed using CBCL tool. Figure 21 shows the relations between Prenatal Exposome (left-hand panel) and child CBCL internalizing problems. Passive smoking during pregnancy is strongly associated with more problems. In the right-hand panel of Figure 21, we show the association between the Postnatal Exposome and internalizing problems. Sleep duration reduces such problems but PCBs also were associated with a reduction in internalizing problems. Figure 22 reports associations with externalizing problems. The Prenatal exposome shows strong associations with active and passive smoking during pregnancy and, at postnatal periods, it shows that healthy diet is protectively associated with child problems but readymade foods, sweets and indoor air pollution are associated with more eternalizing problems.
The prenatal exposome did not show any strong association with cognition based on Raven test (Figure 23), with facility density and mercury exposure being positively associated with the test. The postnatal exposome showed organic food positively associated with the raven scores and fast food and house crowding showed associations in the opposite direction (Figure 23, right-hand panel).
Blood pressure: Results of the analyses linking the prenatal and postnatal exposomes to child blood pressure are presented in Table 12 and Table 13. After p-value correction, significant decrease in systolic BP was observed with markers of the built environment during pregnancy (e.g. facilities density) and with child concentration of some organochlorine compounds (e.g. DDE). Other findings (p<0.05) that did not remain significant after p-value correction could be mentioned: 1) prenatal exposure to bisphenol-A was associated with an increase in both systolic and diastolic BP, 2) low and high fish consumption during pregnancy were associated with higher systolic BP, 3) outdoor temperature and vitamin-D UV dose the day of BP measurement were associated with a decrease in diastolic BP, 4) child concentration of copper and exposure to benzene was associated with an increase in diastolic, 5) Postnatal exposure to benzene was associated with increase in systolic BP, and 6) child concentration of phthalates were associated with a decrease in both systolic and diastolic BP. Results from the DSA selections were concordant with EWAS analyses, with a higher parsimony regarding the selection of exposure within family. Also, DSA selected additional exposures, which taken individually were not associated with blood pressure (i.e. maternal cotinine and child PFOA confounded by a co-exposure). When the prenatal and postnatal exposures selected by DSA were jointly included in a multivariate linear model (Table 14), coefficients differed slightly from those of the univariate models.
Conclusions
HELIX enabled important methodological progress in the study of the Exposome-health relations, and provided one of the first real implementation of an analysis of the potential effects of the early-life Exposome on several components of children health. From a methodological point of view, the EWAS approach used in some former Exposome-health studies was shown to heavily suffer from lack of control for false positive rate. Other more efficient statistical approaches were identified, some of which allowing to characterize interactions between exposure with no cost on statistical power. Our study is one of the largest simultaneously considering over 70 environmental exposures for effects on several components of children health. Compared to former (repeated) single exposure studies, our approach allowed making all tests (usually done in successive studies) explicit, correcting for confounding by co-exposures; we also made attempts in some analyses to correct for exposure misclassification and considering interactions between exposures.
Associations of BMI with postnatal (but not prenatal) levels of Persistent Organic Pollutants need to be considered with caution given the potential for reverse causation. The prospective nature of HELIX is a clear strength given the increased potential for such bias in cross-sectional analyses. By simultaneously testing a large number of exposure factors, the Exposome approach allows to discard confounding by co-exposures and explicitly account for multiple testing, paving the way for a better characterization of the overall influence of environmental factors on human health.
WP6 - Environmental Burden of Childhood Disease and Health App
WP6 aims to estimate the burden of common childhood diseases (obesity, asthma and neurodevelopment) that may be attributed to multiple environmental exposures in Europe. Specific objectives for the full project were:
▪ To construct scenarios for a health impact assessment;
▪ To obtain exposure estimates;
▪ To obtain exposure response data;
▪ To obtain burden of disease estimates;
WP6 first developed an evaluation of the environmental burden of childhood disease in European Union (28 countries). In addition, scenarios for health impact assessment (HIA) have been quantify on active transportation in children, that will integrate multiple exposures (e.g. air pollution, physical activity, traffic incidents) and outcomes (e.g. mortality, morbidity, etc.). The HIA was based on the exposure and outcome data obtained from WP 1-5. In addition, the HIA work includes the design and development of a quantitative model to estimate the risks and benefits of the active transportation in children.
Environmental burden of childhood disease
In this study we aim to estimate the burden of childhood disease due to environmental risk factors in the European Union (EU) of the 28 countries, describing the impact of six environmental exposures (particulate matter less than 10 micrometer of diameter (PM10) and less than 2.5 micrometer of diameter (PM2.5) ozone, secondhand smoke, dampness, lead and formaldehyde), identifying priorities in environmental health policies for childhood in Europe, and highlighting research and risk management necessities.
Selection of environmental risks and health outcomes
This burden of disease was focused on environmental risk factors in children. Metabolic and behavioral risk factors (e.g. sedentary, nutrition, active smoking), and infectious diseases, were excluded from the assessment. The selection of environmental risks and health outcomes were based on the following criteria: evidence for a causal relationship between exposure to the environmental risk factor and the health effect (based on meta-analyses, World Health Organization (WHO) guidelines, or previous risk assessments), independent health effects between the risk factors, availability of exposure data at national level, and availability of baseline health statistics at national level.
Data Collection
Our study considered the population between 0 to 18 years old of the 28 European Union Countries (EU28) (Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxemburg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden and United Kingdom). Population data by country and age were collected from Eurostat database and Institute for Health Metrics and Evaluation (IHME).
Health data was collected from the Institute for Health Metrics and Evaluation (IHME) and the World Health Organization for asthma, mild mental retardation, otitis media, lower respiratory tract infections and infant mortality. From scientific papers and reports, data for cough and low respiratory tract symptoms were collected. Exposure data were collected from IHME for lead, PM2.5 ozone and secondhand smoke, and Environmental and Health Information System (ENHIS) for dampness.
Environmental Burden of Disease
The environmental burden of disease analysis was performed following the Comparative Risk Assessment Approach proposed by the World Health Organization, and the global burden of disease project. The environmental burden of childhood disease was estimated for exposures above defined thresholds, if any, based on a counterfactual exposure distribution that would result in the lowest population risk. The feasibility of reaching the counterfactual exposure levels in practice was not assessed in this proposal. The burden of disease was estimated using the exposure data and a relative risk (RR) derived from epidemiological studies to estimate the population attributable fraction (PAF). This analysis was applied to each exposure-outcome pair. The PAF is defined as the proportional reduction in disease or death that would occur if exposure to the risk factor were reduced to the counterfactual. The burden of disease was calculated using disability weights (DW) and estimates of duration (L) of each outcome.
See below the explanation using three general formulas:
(1) PAF = f x (RR-1 ) / f x (RR-1)+1
(2) AP = PAF x P
(3) DALY = AI x DW x L
Where: PAF: Population attributable fraction; f: Fraction of population exposed; RR: Relative risk; AP: Attributable prevalence; P: Background prevalence; DALY: disability-adjusted life year; DW: Disability weights; L: Duration of condition.
The disability weights used in this analysis were those proposed by the global burden of disease project or used in previous environmental burden of disease estimations. For some cases an approach based on the unit of risk (UR) was used. The UR estimated the absolute number of cases that are to be expected at a certain exposure, and then was transformed to DALYs using disability weights and the duration of the condition. Burden from lead and mental retardation, PM2.5 and low respiratory infections, secondhand smoke and low respiratory infections and otitis media were obtained from the global burden of disease project 2015.
Environmental Burden of Disease in the EU of 28 countries.
Seven different exposures where identified under the inclusion criteria, associated to six different health outcomes. We estimated that in the population aged below 18 years of the EU28, the seven environmental exposures (lead, PM10, PM2.5 ozone, secondhand smoke, dampness and formaldehyde), are responsible for 210777 disability adjusted life years (DALYs) annually. Fifty nine percent of these DALYs were attributable to particulate matter (PM10 and PM2.5) exposure, 20% to secondhand smoke, 11% to ozone, 6% to dampness, 3% to lead, and 0.2% related to formaldehyde.
Particulate matter less than 10 micrometer of diameter (PM10) and less than 2.5 micrometer of diameter (PM2.5).
PM10 is associated with infant mortality (< 1 year old)(WHO/Europe 2013) and asthma (5-18 years old). Of these, infant mortality was associated with the major burden (93147 DALYs annually), followed by asthma (13904 DALYs annually). PM2.5 is associated with low respiratory infections (< 18 years old) and was estimated to produce 134032 DALYs annually.
Secondhand smoke.
Secondhand smoke is associated with asthma (< 14 years old), low respiratory infections (< 5 years old) and otitis media (< 5 years old). Of these, asthma was the disease with the mayor burden, resulting in 20880 DALYs annually, followed by low respiratory infections with 9728 DALYs annually, and Otitis media with 2062 DALYs annually.
Ozone.
Ozone is associated with low respiratory symptoms (including cough)(5-14 years old). Cough days related with ozone was estimated to result in 10057 DALYs annually. Other days with low respiratory symptoms related to ozone were estimated to result in 14122 DALYs annually.
Dampness.
Dampness is associated with asthma in children less than 14 years old, this disease was estimated to result in 12954 DALYs annually.
Lead.
Lead exposure is associated with mild mental retardation (< 14 years old), this disease was estimated to result in 6216 DALYs annually.
Formaldehyde.
Formaldehyde is associated with asthma in children less than 3 years old, and resulted in 33 DALYs annually.
Sensitivity analysis.
Sensitivity analysis for PM10 and asthma assuming a counterfactual of 1.9 μg/m3 resulted in 18681 DALYs, and assuming counterfactual of 20 μg/m3 resulted in 3885 DALYs. We also performed a sensitivity analysis using a different exposure-response function between PM10 and asthma from new meta-analysis; in this analysis was estimated 45098 DALYs. The sensitivity analysis of PM10 and infant mortality assuming a counterfactual of 1.9 μg/m3 resulted in 124794 DALYs, and assuming counterfactual of 20 μg/m3 resulted in 30499 DALYs. Sensitivity analysis for secondhand smoke assuming the minimum percentage of secondhand smoke reported in the European Union 28, resulted in 12848 DALYs for asthma, 6867 DALYs for low respiratory infections, and 1168 DALYs for otitis media. Sensitivity analysis for dampness and asthma, using mould (instead of dampness) as an exposure resulted in 11470 DALYs. Finally the sensitivity analysis for formaldehyde and asthma using a 60 μg/m3 as a threshold resulted in 1667 DALYs.
Development of the DigiBIB APP: linking environmental exposures and electronic health records.
Using learning from earlier work-packages we have developed a multi-functional prototype smartphone APP which can track users’ locations and record research calibre real time assessments of health and wellbeing, with the potential to calculate personalised exposure to air pollution and other exposures. The APP also allows users to access their own electronic health record, and will allow dyadic communication between health professionals and participants (for example, along medical professionals to gain contextual information about levels of pollution users are experiencing, and allowing users to report health concerns such as wheezing and asthma). The APP is currently being piloted within the Born in Bradford cohort study where we will explore acceptability and usage amongst different socio-economic and ethnic groups.
Conclusions
Our results show a large impact of environmental exposures in child health across Europe (Figure 24, Annex I). This study found that the environmental risk factor for child health in the EU28 with the largest impact was air pollution (PM10, PM2.5 and ozone) exposure, representing more than two thirds of the environmental burden of disease of the seven exposures combined. This is similar to a previous burden of disease study in six countries of Europe. Secondhand smoke also showed a large impact, resulting in 20% of the environmental burden of disease in European children.
The assessment of environmental burden of disease estimates offers the opportunity to identify priorities and solutions for policy and research. These priorities and solutions are presented in this study as recommendations for authorities, public health specialist and researchers, and are as follows. First, this study found a real need to create and implement effective policies to reduce children exposure to environmental risk factors across Europe, with special attention to major risk factors such as air pollution and secondhand smoke. Second, there is a need to create common European databases, which collect and harmonize exposure data for (old and new) environmental risk factors, especially for children. Third, the need for epidemiological studies on multiple environmental risk factors, with special attention to provide dose-response functions, with harmonized exposure and outcome definitions. Currently, a number of European research projects focus on the “exposome” concept as HELIX, EXPOsOMICs and HEALS (HEALS 2017), and some of them collected relevant information on multiple exposures and outcomes in children. These projects should be seen as a good examples to start the filling some of these gaps (Table 17, Annex I).
Potential Impact:
Key messages and implications
Novel Tools for Individual Exposures
▪ Helix has demonstrated that it is possible to perform a harmonized, though extensive, fieldwork using the same protocols for sample collection and clinical examination in six European mother-child cohorts.
▪ Children across Europe are exposed to a wide range of environmental chemicals in fetal life and childhood, and the exposure varies substantially between countries and by compounds.
▪ The concentrations found in the maternal samples were in general higher than concentrations measured in the child samples.
▪ For most of the persistent compounds the correlation between maternal and child levels was high, while considerably lower correlations were observed for most non-persistent compounds.
▪ There is a substantial variability in exposure during pregnancy and over a year in school-age children for the majority of the non-persistent compounds.
Novel Tools for Outdoor Exposures
▪ HELIX developed and applied standardized exposure assessments for urban exposures including built environment indicators, air pollution, noise, green space and temperature and successfully applied these to different study areas.
▪ Urban exposures vary substantially within and between children and pregnant women and within and between cities.
▪ Generally, correlations between the different urban exposures are weak to moderate.
▪ A considerable proportion of children and pregnant women are exposed to levels of urban exposures that are above international guidelines.
Integrating Multiple Exposures and Uncertainties
▪ The exposome is high-dimensional, as it cannot be reduced to a small set of principal components.
▪ Although correlations within the same exposure family can be high, correlations between exposures from different families were low. This supports that epidemiological studies focusing on a reduced set of exposures may not be confounded by having omitted exposures from other families.
▪ Clustering of exposures may not be adequate for multicentre analyses on the effects of the exposome on health outcomes as location is a strong determinant of one’s personal exposome.
▪ PBPK models can be used to integrate more individual information (e.g. breastfeeding) together with biomarker measurements in order to rebuild realistic exposure scenarios.
▪ For most non-persistent chemicals, multiple pools of multiple urines would be needed to obtain excellent reliability in exposure assessment; for example, 4 pools of 15-20 urines each would be needed to get a reliable estimate over the entire pregnancy. This is important for the design of future biomarker studies and for measurement-error corrections in HELIX.
▪ The variability in environmental outdoor exposures was mainly determined by the activity patterns of participants’ daily life rather than by individual or city characteristics. These results confirm the need of using personal exposure assessment methods for outdoor exposures in epidemiological studies and the importance of having access to accurate, inexpensive and not burdensome personal exposure tools.
▪ Future exposome studies should continue refining exposure assessment through repeated collection of biospecimens and personal dosimeters in even larger populations.
Integrating Molecular Exposure Signatures
▪ In the HELIX project we have successfully generated a unique molecular profile data resource for exposome research comprising of urinary and serum metabolomics, plasma proteomics, blood cell DNA methylation, transcriptomics and miRNAdata for 874 children in the HELIX subcohort, with up to 1198 children’s samples analysed for individual omics platforms.
▪ An EXposure Wide Association Study (ExWAS)conducted with metabolomics, proteomics and the child-matched full exposome dataset from HELIX identified several significant clusters of associations.
▪ Serum polyunsaturated glycerophospholipids were associated with exposure to heavy metals and PFASs, also urinarytrimethylamine-N-oxide with exposure to heavy metals and urinary hippurate and proline betaine with OPs. These signatures are likely to represent common routes of exposure, e.g. fish and seafood as a common source of polyunsaturated fatty acids and metals.
▪ We also found associations between OCs and PBDEs with plasma adipokines, which can reflect a link with fat mass, e.g. fat mass as a driver of adipokine expression and storage of lipophilic chemicals.
▪ We have defined the major determinants of the metabolome in European children and identified a novel link between threonine catabolism and BMI.
Linking the Exposome to Child Health
▪ Simulation studies indicated that in the presence of correlation between exposure factors, Exposome wide association studies, even correcting for multiple testing, are expected to suffer from a high rate of false positives.
▪ Other statistical approaches, such as the Deletion/Substitution/Addition (DSA) or Elastic Net algorithm, have a lower false detection rate, at a cost of a decreased sensitivity for a given sample size. The efficiency of several approaches aiming at identifying statistical interactions between exposure factors has also been characterized.
▪ Studies linking the Exposome to various component of children health (birth weight, postnatal growth, respiratory health and allergy, blood pressure, neurodevelopment) have been conducted in 1300 children, as well as a study linking the urban Exposome to birth weight in 30,000 children.
▪ The Exposome health associations observed in Helix include a relation between lead and green space exposure and birth weight; relations between active and passive smoking (prenatally), indoor particulate matter, copper (postnatally) and body mass index; facility density (prenatally), DDE and HCB (postnatally) and systolic blood pressure; passive smoking (prenatally), PCB and sleep (postnatally) and internalizing problems.
▪ By simultaneously testing a large number of exposure factors, the Exposome approach allows to discard confounding by co-exposures and explicitly account for multiple testing, paving the way for a better characterization of the overall influence of environmental factors on human health.
Environmental Burden of Childhood Disease
▪ HELIX has estimated the childhood environmental burden of disease in Europe, highlighting the relevance of environmental factors in children's health across the European Union.
▪ HELIX has developed new evidence that could be integrated into future health impact assessments to estimate more comprehensively the environmental burden in childhood health and to assess future policy interventions.
▪ HELIX has also highlighted the need for more evidence on the Exposome that can be translated into evidence-based policymaking to protect and improve public health.
Use and Dissemination of results
HELIX set out to produce more robust evidence base on links between exposome and human health and well-being. The HELIX project produces a lot of data and hence results on exposome research. As a reminder the objectives of the HELIX project are reiterated as follows:
• To measure a range of chemical and physical environmental hazards in food, consumer products, water, air, noise, and the built environment, in pre- and post-natal early-life periods;
• To define multiple exposure patterns and individual exposure variability (temporal, behavioural, toxicokinetic);
• To quantify uncertainty in the exposure estimates;
• To determine molecular profiles and biological pathways associated with multiple exposures;
• To obtain exposure-response estimates for multiple exposures and child health outcomes;
• To estimate the burden of childhood disease in Europe due to multiple environmental exposures;
• To strengthen the knowledge base for European policy.
Through these objectives and associated methods, a significant amount of data has been and will be produced and interpreted. This will lead to conclusions that can be used in health/environment/research policies. The results can help us to plan ways of improving public health. Moreover, the results can be useful for policy makers in the domain of environment and health and even in the public health domain. The results of the project will also point to further open questions which can lead to more research.
The project has delivered a wide range of results. These are produced in different formats: Scientific reports; scientific articles; guidelines and recommendations for stakeholders or professional practitioners; dissemination at a range of meetings.
The products in short are:
O Scientific article
O Newsletters
O Brochure
O Press releases
O News articles
O Conference/meeting attendance (scientific, advisory board, policy)
O Summary future research needed
O Section implementation in paper
O Social media/media
Each of these formats have their own value in dissemination of the results of the project.
The results in HELIX have been produced in seven work packages (WP). Each WP has produced its own reports with results, as well as articles in peer reviewed journals. There are different categories of stakeholder groups that are addressed in HELIX.
In order to ensure stakeholder involvement, we have used the stakeholder network of the HELIX project. This task involved establishing and maintaining structural and ad hoc contacts with policy makers, regulators, research networks and other stakeholders. At the start of the project a Stakeholders Forum was suggested across the different Exposome projects. This has not been realized yet, but a proposal for a network has been filed as a COST Action at the EU. An inventory of stakeholders' needs is the starting point of stakeholder involvement. There is a need in Europe for an Exposome toolbox as an important source of information in the decision making process in the domain of environment and health.
These stakeholder groups have been identified within the consortium by the partners, in collaboration with stakeholders attending the different meetings within HELIX.
Scientific reports: Results based on reports of the different WPs aim at scientists and policymakers. These reports have been or will be distributed by the work packages and disseminated via the HELIX website. Other stakeholders can also access these reports.
Scientific articles: The scientific articles have been produced by the members in the HELIX consortium (Table 18, Annex I). Most articles have been published in peer reviewed journals. Even more articles have been submitted and are currently being drafted (Table 19, Annex I). The abstracts of these articles are provided in Deliverable D7.11
General public: The general public has not been a prominent stakeholder so far for the consortium to disseminate the results, however, the partners have presented the project at public events throughout the project. The topic of HELIX is complicated and not yet developed into full strength in order to disseminate the results in an easy to understand manner.
Throughout the project, partners have contributed to a significant number of (newspaper) interviews and video reports in the popular media. The general dissemination about the concept of the HELIX project to the general public has taken place in different countries. In addition, video messages were produced and disseminated via the project website (http://www.projecthelix.eu/).
Webinars: The project established a collaboration with the Exposure Science and the Exposome Webinar Series initiative of the National Institute of Environmental Health Sciences (NIEHS, USA). The first HELIX Webinar was recorded in October 2014, followed by a joint Stakeholder Webinar with the EXPOsOMICS project. The last combined webinar was held in December 2015 as joint session with NIEHS and Exposomics. We intend to keep collaborating on this type of engagement, where possible with other initiatives.
Stakeholder interaction: The interaction with stakeholders outside the consortium has been analysed in HELIX Dissemination Strategy (D7.10). Furthermore, the most favorable engagement with stakeholders has been described.
The results of the research of HELIX are important for different stakeholders. Results of research projects do have more value if they are translated into practical and understandable recommendations. This process is not always easy. Results from research always raise more questions. However, each result will point in a certain direction. If some health benefit can be added to a result out of the HELIX research it should be mentioned.
The presented results of the different studies within HELIX all have some value for implementation. Science produces a lot of results which can lead to more scientific questions. But there are always results that can be used to improve our world. Sometimes these results are still very indicative. Sometimes results are not causal to explain mechanisms. But always there are possible conclusions that cause no harm to take action upon.
A result can be descriptive and confirm common knowledge. There are some items that have been studied that are a proxy for a range of factors that cannot be described.
The results of the research of HELIX are important for different stakeholders. Results of research projects do have more value if they are translated into practical and understandable recommendations. This process is not always easy. Results from research always raise more questions. However, each result will point in a certain direction. If some health benefit can be added to a result out of the HELIX research it should be mentioned.
To enhance the potential impact of the project and improve dissemination at all levels, HELIX has released a set of recommendations. These are formulated based on the scientific field work and on literature analysis.
Accessible Data Inventory
As part of the remit, HELIX has constructed the HELIX Data Warehouse, containing HELIX Foreground data (i.e. data generated by the HELIX study) and participating cohort Background data (i.e. data originally generated by the individual cohorts). The data is available for research purposes to researchers outside the Consortium for work on specified manuscripts. A clear overview of the inventory and included variables is available from www.projecthelix.eu/index.php/es/data-inventory. Interested parties find here the request protocol has been established to guarantee adherence to cohort data access rights, data protection regulations, and ethics approvals relevant to the cohorts participating in HELIX.
Researchers external to the HELIX Consortium who have an interest in using data held in the HELIX warehouse for research purposes can apply for access to data for a specific manuscript at the time. The analysis of these data will be considered outside the remit of the HELIX project. The applicant has to be affiliated with an institution with competence in conducting research projects and which has agreed to be responsible for the conduct of the proposed manuscript. Junior researchers must have a scientific supervisor belonging to such an institution. All proposed manuscripts must have a principal investigator with scientific responsibility for the project. For each accepted proposal, a data transfer agreement (DTA) will be signed between the cohorts participating in HELIX, and the receiving institution. A detailed description of the Data Warehouse can be found in Deliverable 5.4 submitted to the EC following the official end of the project.
Further procedures to for the request of biological materials available from the participating cohorts will be finalized and published during the course of 2018.
Policy implications
The HELIX project has produced different products. The project is finished at a stage were most results of individual studies are analysed and made ready for publication. The final scientific symposium produced a limited amount of suggestions of policy and societal implications of the results of the studies.
The complicated exposure scenarios can lead to large exposure mis-classification problems. However, the combination of PBPK modelling and realistic exposure data can contribute to rebuild realistic exposure scenarios and reduce uncertainty. This approach can lead to the identification of very sensitive individuals. Results have shown this for PBPK modelling of PFOA and PFOS.
Another implication of the studies shows that intervention early life exposure can contribute to decrease comorbidities in adulthood. Therefore we need standards based on vulnerable time periods.
There are several individual studies that have analysed health risks related to multiple exposures in a novel systematic way and these show possible implications for health. These are related e.g. with environmental exposures and child blood pressure, endocrine disruptors and lung functions, organic compounds and ADHD or social competence, environmental exposures and obesity.
These findings show that we should develop:
▪ Governmental policies to reduce environmental risk factors
▪ Early prevention in children that should be effective (children with elevated BP are more at risk to be hypertensive in adulthood)
▪ Measures to improve quality of life and decrease health costs
Ultimately, results would help us to guide public health efforts by allowing us to intervene on those chemical agents or urban exposures that are most likely to be associated with childhood diseases. Results would thus help prioritisation of interventions.
Evidence-based knowledge translation
The scientific evidence created by HELIX will help to disentangle the relationship and impact of multiple exposures and health outcomes. To translate this scientific evidence into decision-making processes will require the development of tools like Health Impact Assessment. During the HELIX project, a health impact assessment approach was used to estimate the health impacts of seven environmental exposures in the European Union (EU28). This exercise helped to identify the main priorities (based on evidence) between these seven environmental exposures individually in the EU28. Applying a health impact assessment approach using the Exposome concept instead of using individual exposures will require more scientific evidence, especially from epidemiological studies, and the development of further approaches to translate the Exposome concept to a more integrated unit.
Until now the evidence around Exposome is not yet integrated into a simple and harmonize unit, in order to be translated to policies and design interventions. For that reason, in the future, the development of a simple approach to translating the Exposome concept into policies and interventions will be required. A possible way to tackle this would be the development of an index, which integrates different dimensions and characteristics of the Exposome and could be used to communicate the health risks and benefits of multiple exposures in a single unit. The “Exposome Index” could be an easy tool to identify where and when a policy or intervention will be required, will also help to prioritize interventions, compare populations, regions, and develop trends. Finally the “Exposome Index” could also help to assess the efficacy of policies and interventions, based on their performance to improve health determinants and health outcomes. Further research will be needed to develop the “Exposome Index” to the available evidence around the Exposome concept; the potential to develop an “Exposome Index” could simplify the translation of the Exposome concept (and evidence) into policy.
Continuation
During the course of the project, the objective was to cluster coordination activities between the HELIX WPs. The focus of this activity concerned HELIX WP7 (Dissemination and Engagement) and WP8 (Management) intent to exploit outcomes of the project most efficiently.
Amongst other, this concerned the development of a dissemination strategy that included the identification of stakeholders, identification of appropriate dissemination frameworks and methods, with exchange between and coordination of, appropriate initiatives. This is outlined in Deliverable 7.10 Final Dissemination Strategy.
These recommendations for future dissemination are depending on the progress of the analysis of the collected data in HELIX. The plan is to continue to publish scientific papers after the official end of the project. It will take at least another year before most of the papers are published. All the researchers within the consortium are going to fill in the templates about the results of their scientific efforts.
The compilation of all of the templates will be added to the website of the project. Thus, the information will be available for the stakeholders. With help of the provided templates it will be more convenient to have an overview of all the results. In addition, the researchers are asked to add any created visuals to bring across their message, by using Infographics, shared presentations / videos on YouTube or SlideShare.
The HELIX website (http://www.projecthelix.eu/) will all add the connection to future studies, new technologies etc. based on the samples collected during the project (continuation process). The website and HELIX social media accounts - twitter (@greenhealth4eu), Facebook and LinkedIn - will continue to be active and function as reference sources and as tools for viral messaging.
In the long-run, exploitation of the HELIX results and the knowledge acquired during the HELIX project will be achieved via involvement of partners in other international projects, including those funded by the EU (Table 20, Annex I). HELIX partners are already involved in mayor H2020 projects including LifeCycle (Early-life stressors and LifeCycle health) and HMB4EU (Science and policy for a healthy future), as well as the newly proposed STOP (Science and Technology in childhood Obesity Policy). Further exploration of the gaps in exposome research and use of HELIX data is expected to lead to a significant number of new proposals in the near future. Furthermore, the involvement of partners in policy networks at the national and international level, including different Directorates of the EU, Member States’ health ministries and their commissions, and regional health authorities, will facilitate the awareness of HELIX results and their possible use for the development and implementation of policies related to public health and environment.
Plan for continued dissemination of project results
Webinars
The project established a collaboration with the Exposure Science and the Exposome Webinar Series initiative of the National Institute of Environmental Health Sciences (NIEHS, USA). The first HELIX Webinar was recorded in October 2014, followed by a joint Stakeholder Webinar with the EXPOsOMICS project. The last combined webinar was held in December 2015 as joint session with NIEHS and Exposomics. We intend to keep collaborating on this type of engagement, where possible with other initiatives.
Brief
A short brief on what HELIX does and what ‘exposome’is has been produced for dissemination purposes (Table 21, Annex I). This brief is available on the website and will be used in communication about ‘exposome’.
Newsletter
Electronic newsletters send through the earlier mentioned Mailchimp system have been published at regular intervals during the project. The news alerts aim at alerting users to key developments and headline results, events, and personnel involved in the project, and to provide links to other related activities (e.g. parallel studies). It also provides brief commentaries on the implications of these developments for policy, and invites similar commentaries and reviews from users or other researchers. We will continue this effort in order to update stakeholders of ongoing developments and publications. This resource also allows for new stakeholder to sign up as they see appropriate, providing for the dynamic changes of people entering and leaving the field of interest.
Workshops
Efforts are made to find financial and time resources to organise HELIX-exposome focussed meetings beyond 2017. The earlier mentioned COST Action, is a potential way to organise workshops to a network of exposome researchers. Also workshops on Exposome can be organised in different scientific conferences in 2018 e.g. PPTOX VI and ISEE 2018.
Policy impact
The department of Policy and Global Development is the core element in ISGlobal's knowledge transfer strategy. The department's dual function as a think tank and a catalyst for ideas and action embodies the institute's strategy of studying real world problems to effect change. The HELIX management is collaborating with this department in its efforts to translate the scientific results into clear recommendations for policy advisors and decision makers. Several partners within the HELIX consortium have discussed the topic of policy transfer of Exposome research results. This Policy Core Group has made a work plan on how to deal with science policy transfer. The set-up of the final scientific HELIX conference has been built upon insights of the workplan.
Further resources
Publication of materials for the non-specialist audience is a tool within the project. Moreover, the use of illustrations, tools and further media developed within the project will be actively encouraged. This can be incorporated in the dissemination of the templates as they are proposed in deliverable 7.11 – the report on awareness and societal implications of HELIX research and development.
A stand-alone document including the project description, abstract, the main policy relevant questions that were addressed, and first results, will be prepared in lay language.
List of Websites:
www.projecthelix.eu
