The National Institute of Environmental Health Sciences has long-supported the concept of the exposome, funding studies that are now being defined as exposomic in nature. The Institute recently established a strategic goal of transforming exposure science and has identified the exposome as a possible approach (19). They plan to advance characterizations of environmental exposure assessment at both the individual and population levels which will be accomplished through tools and technologies for multi-scale measurements.

Recently, the National Institute of Environmental Health Sciences funded the Health and Exposome Research Center: Understanding Lifetime Exposures (HERCULES) project at Emory University (http://emoryhercules.com/) with the stated goal of understanding lifetime exposures. The main aims of HERCULES are: 1) to provide greater access to exposome-related approaches such as systems biology, metabolomics, high throughput toxicology, spatial and temporal statistical models; 2) to facilitate communication of the importance of environmental factors in disease using exposome principles; and 3) to expedite translation of novel scientific findings to develop novel sustainability, prevention or treatment strategies in humans.

Omic biomarkers are a class of biomarkers of current scientific interest (Table 2). The discovery and use of these biomarkers have increased rapidly, due to the advent of high-throughput technologies and innovations that include improved sample preparation, robotic sample-delivery systems, automated data processing, and use of multivariate statistical methods, with associated reductions in cost. Investigators have begun to use these biomarkers in larger-scale population studies of the exposome.

One of the greatest challenges with these studies is applying omics technologies to generate meaningful results. Epidemiologists must strive to understand the principles of omics and determine when it is appropriate to include biomarkers identified using these technologies. In addition, no single omics approach will suffice to characterize the exposome, and integration of omics outcomes and other sources of exposure information will be needed to deepen our understanding of the causes of disease. For example, metabolomics studies have shown links between gut flora, diet and cardiovascular disease (24, 25). Omics technologies have shown utility in determining toxicity and mode of action in risk assessments and assessing the health impact of an exposure using analysis of variance (ANOVA) approaches together with pair-wise comparisons between dose groups and the corresponding control (26). Thus, what might be considered an advance in exposomics can equally be considered a natural combining and extension of traditional tools.

Many of the techniques in assessing exposure and the exposome involve molecular epidemiologic studies. Molecular epidemiology is the use of biological markers (exposure, effect, susceptibility) in epidemiologic research (27, 28). In assessing the exposome, biomarkers of exposure may be the most useful. The development of adductomics, which measure the full complement of protein adducts might be useful in improving exposure assessment in epidemiologic studies because of the ability to reflect extended exposure (28, 29). As some omics applications mature and systems biology becomes more incorporated in molecular epidemiology, the understanding of the exposome will be increased, and it should be possible to more broadly explore exposure-disease relationships, to study effect modifiers, and to obtain insight into temporal and multi-level factors in health and disease.

Remote sensing is a key innovation in exposure science and is defined as the measurement of some property of a phenomenon, object, or material that is not in direct physical contact with the population being studied (23). The use of sensors for not only remote monitoring but personal monitoring is growing. Sensors are now used to measure clinical parameters such as blood pressure and glucose levels, and new sensors are being developed to measure biomarkers, such as portable adhesive sweat analyzers (30). The devices—with remote-sensing-based spatial referencing technologies and modeling—enable the continuous, real-time, real-world assessment of exposures that vary according to an individual’s location, activity, and lifestyle. The use of smart phones and tablet technology is only likely to grow and provide enhanced opportunities to collect exposure information (23, 31). Apple has created ResearchKit (https://developer.apple.com/researchkit/) a framework for Apple platforms that can be used in medical research of diseases such as Parkinson’s, asthma, diabetes, heart disease and breast cancer.

In another example used in a current exposomic study, the ExpoApp, a mobile application created specifically for the HELIX project, uses a Global Positioning System (GPS) and a built-in accelerometer to track a person’s location and measure physical activity every 10 seconds (32). Participants in the HELIX project will wear ExpoApp-enabled smartphones for a week, along with air pollution and ultraviolet radiation monitors, and the data will be used to calculate the amount of air inhaled and an individual’s exposure to air pollutants. A challenge is that individuals with access to information from such applications may alter their behavior and thereby inject biases into the study data.

Geographic Information Systems are informative for tracking the acquisition, editing, analysis, storage, and visualization of geographic data (33). Geographic Information Systems allow for mapping a variety of data, for example environmental, topographical or health-related to understand trends and patterns. Information from Geographic Information Systems would fall into and provide clarity to exposures in the general external domain of the three domains of the exposome described by Wild (3). Geographic Information Systems have been reported to enhance exposure assessment in epidemiologic studies, because such systems can provide information on broad environmental contaminant levels maps or to define a population (34). For example, the Research Center on Health Disparities, Equity, and the Exposome (http://rchdee.uthsc.edu/) maintains the Public Health Exposome Data Information System, a longitudinal, 30-year data set that integrates over 20,000 environmental and health data records (9). The database contains data indicators of the atmosphere, climate, water, land cover and land use that have been geocoded. In addition, a considerable amount of other data are geocoded and included in this database, derived from ground stations and field samples of emissions, heavy metals, toxic dump and storage sites, and Brownfields collected by the Environmental Protection Agency with other federal, state, and local agencies.

Survey instruments gather information for which there is no way to quantitatively measure exposures and have long been used in epidemiologic studies. These tools may be important in documenting retrospective exposures, which currently is not feasible with other exposomic tools. Modeling the data to improve exposure matrices will be an important aspect of using exposomic data.

With the advances in molecular medicine and development of omics technologies, the field of biomedical informatics has evolved as a discipline. Genome-wide association studies (GWAS) and studies combining genomic and phenotypic data have required the development of new biostatistical methods for quality control, imputation, and analysis issues such as multiple hypothesis testing. Developments in exposome science have revealed the need for evaluating the interrelationships among phenotype, genotype, and exposure data. As shown in the different approaches discussed here, these data will be heterogeneous, wide-ranging, and massive; will frequently involve time series; and will require high velocity processing (35). Some efforts have been made for the analysis of EWAS, but new research is required to develop approaches for data management, analysis and visualization.

The term “reality mining” refers to the analysis of behavioral and self-reported data extracted from social networks and other portable devices’ applications (36). By continually logging and time-stamping information about a subject’s activity, location, and proximity to other users, it is currently possible to identify patterns in the data and translate them into maps of social relationships. By definition of the exposome, all exposures are to be taken into account. Logistically, that is impossible but by using social media networks or citizen scientists’ efforts, such as HabitatMap (http://habitatmap.org/), additional exposure information could be obtained. This relational information could have much broader implications including the improvement of existing computational models of exposure, disease status of an individual, and disease spread (37). A recent report demonstrated how the mobile application, Yelp, was used to identify foodborne illnesses in New York City restaurants (38).

In an EWAS employing techniques partially adapted from GWAS, Patel et al. (39, 40) used chemical, clinical and questionnaire data from the National Health and Nutrition Examination Survey (NHANES) cohorts to evaluate associations between multiple environmental factors and type 2 diabetes mellitus and serum lipid levels, respectively. The study followed two methodological steps analogous to those in a GWAS. First, the authors considered a panel of environmental assays, and identified those significantly associated or correlated with diabetes or serum lipids while controlling for multiple confounders. Second, they validated the associations by testing significant findings in other NHANES independent cohorts. The authors identified environmental factors, including select chemicals, corroborating earlier findings. Additionally, Patel et al. (41) screened for gene-environment interactions by integrating results from GWAS and EWAS. Properly designed, EWAS studies can lead to discovery of biomarkers for exposure and disease and establish a molecular basis for the cause of environmental diseases (42). Patel et al. (43) described the “exposome globe” [e.g. Figure 4], which is a visual depiction of the network of replicated correlations between individual exposures of the exposome. The exposome globe allows visualization of clusters of exposure.

Although omics technologies are at the forefront in studies of the exposome and associated biomarkers, omic-based measurements do not always reflect exposure and may instead be products of normal cellular function (3). Proteomics, transcriptomics and metabolomics are a few of the omics technologies that have shown great promise for exposomic studies. As our understanding of basic biologic pathways grows, perturbations in the pathways as measured by omics result in improved interpretation with respect to how health is affected. Other types of exposure information of value to EWAS include databases that contain exposure information or population demographics data that can impact exposures. Use of remote sensors to gather environmental data may help ascertain exposures to the population as a whole. Personal monitoring using sensors to determine exposures, physiological factors, and geographic location is beginning to be used in environmental health studies to assess the “total” exposure of study participants.

Two strategies have been identified for characterizing the exposome (2). One is a ‘bottom-up’ strategy in which all the exposures in a person’s exposome are measured at set time points. Although this approach would have the advantage of identifying important exposures in the air, water, or diet, it is not currently feasible and would miss essential components of the internal chemical environment due to such factors as gender, obesity, inflammation, and stress (44). At the other extreme is to measure a combination of omic endpoints and legacy biomarkers in repeated blood specimens. This strategy has been referred to as ‘top-down exposomics’ (2). This data-driven (or “agnostic”) approach lacks specific hypotheses (2). According to Rappaport (2), the exposome would consist of a profile of exogenous and endogenous exposures, but would not pinpoint their source. Since it is currently not feasible to measure all chemicals in the blood, it has been proposed (44, 45) to focus initially on the most prominent classes of toxicants known to cause disease, namely, reactive electrophiles, endocrine (hormone) disruptors, modulators of immune responses, agents that bind to cellular receptors, and metals. Exposures to these agents can be monitored in the blood either by direct measurement or by looking for their effects on physiological processes (such as receptor-based signaling). These measurements could help generate signatures or profiles for these exposures in the blood. These profiles could be used to help identify key exposures associated with a disease by comparing the profiles between cases of that disease and controls, preferably from longitudinal studies. Additionally, once important profiles of biomarkers have been identified the sources of exposure could be determined and methods to reduce the exposures could be identified. Therefore, discovery-driven research and hypothesis-driven research should be considered complementary and synergistic.

For an exposure to be a cause, the exposure must precede the outcome. Reverse causation is a situation in which the outcome precedes and causes the exposure instead of the other way around. This could occur, for example, if a person moved or changed his/her address as a result of a condition in the domicile that was making him or her sick. It is of particular concern in retrospective and cross-sectional studies. In prospective cohort studies, since exposure is determined in advance of disease onset, the probability of reverse causation is greatly diminished (1). EWAS study designs have been developed that can reduce the possibility of reverse causation (48).

Statistical inference problems with the use of omics methodologies involve the simultaneous test of thousands of null hypotheses. The multitude of comparisons made in these studies will result in both false positive (Type 1 errors) and, if the correction for multiple comparisons is overly conservative or power is inadequate, false negative (Type 2 errors) results (49).

Univariate models consider each predictive variable separately, and their association with the outcome of interest is tested using the same statistical model. The Family Wise Error Rate (FWER) and the False Discovery Rate (FDR) have been used to characterize the number of associations that could falsely be declared statistically significant (50, 51). The Family Wise Error Rate is the probability of making one or more false discoveries, or type I errors, among all the hypotheses. Bonferroni and other correction methods are commonly used; however, these often produce exceedingly conservative thresholds (50).

In an exploratory approach (‘top-down exposomics’, described above), it may be preferable to use a less conservative correction strategy. The False Discovery Rate is the expected proportion of errors among all associations declared statistically significant. To control for the False Discovery Rate, one must define the proportion of positive findings that are allowed to be false, usually 5%. Methods have been developed for both approaches and are described in detail elsewhere (50).

While univariate methods are useful for uncovering simple relationships between predictors and responses, they are also likely to overlook relationships involving combinations of factors. Different multivariate methods have been developed for this purpose. Multivariate analyses aim to summarize the information contained in large datasets into a few synthetic variables [the principal components] that capture the latent structure of the data. Because these methods effectively reduce the number of dimensions necessary to represent the data, they are often referred to as dimension-reduction methods. These methods have been described in detail elsewhere (50). Due to the density of the data in exposomic studies, identification of independent associations will be challenging. Patel and Ioannidis (46, 47) have proposed some agnostic approaches to aid in the analysis of the large number of variables found in exposomic studies. Smith et al. (52) proposed the use of genetic traits and Mendelian randomization techniques to study highly confounded risk factors and disease causation. Such a technique may be useful for exposome studies to investigate the effects of modifiable risk factors of diseases that are too heavily confounded to be studied by conventional approaches.

Variability over time and between subjects is associated with a multitude of intrinsic and extrinsic factors, some known and some unknown. Unlike the genome, omics endpoints are dynamic and likely to show variability in different cells and tissues, and throughout the life of an individual (53). Panel studies in small population samples have been proposed to measure the effect of short-term variability on exposure and omics biomarkers, on individual behaviors (physical activity, mobility, time activity), and on personal and indoor exposures (20).

For exposures with a short biological half-life and little constancy in the underlying exposure behavior, temporal variability may be particularly high. For such exposures, intra-individual compared with inter-individual variability is known to be high, and only repeat measurements over time provide improved exposure estimates (20). Studies to measure daily repeat biomarkers of non-persistent chemicals (phthalates, phenols, organophosphate pesticides) in urine have been proposed to characterize intra- and inter-individual variability in these urine biomarkers, and where possible, correct for the uncertainties in a larger cohort.

With the development of stable and cost-effective high-throughput platforms, large amounts of experimental data are generated. However, data obtained from these platforms are highly sensitive to experimental conditions, and can therefore include considerable noise in the form of measurement error. Statistical methods should be able to estimate technical nuisance variation (50).

Many challenges in measuring the exposome exist. The sheer diversity and variability of exposures individuals experience throughout their lives is impossible to measure. Exposures constantly change in a person’s environment over time (54). Environmental contaminants appear and disappear and an individual may make changes in their lifestyle choices that can increase or decrease exposures. Additionally, the relevance of past exposures remains unknown and the inability to measure past exposures hampers the exposome (54). Certain life stages have been identified as being more susceptible to environmental exposures (i.e., lifecycle “windows of susceptibility”) so relevancy across time may also be an issue.

Biomarkers, one of the primary ways to measure the exposome, have numerous challenges. Their relevance to exposure or their predictive value in non-target tissues may not have been established. The full validation of the biomarker both in the laboratory and the population has not always been performed making the interpretation or relevance of the measurement difficult. Omics technologies have multiple advantages that make them well-suited for exposomic studies but also have limitations in interpretation and validation for exposures and effect of exposures. The management of large datasets is still a challenge for these technologies; and these technologies also require major investments in equipment and expertise.

Other techniques that may be important to measure the exposome also have limitations that include the uncertainty about relevance of the measurement, variability over time, and the inability to measure past exposures. While environmental monitoring (personal or remote), geospatial information, and reality mining may tell us what exposures a person may have had, the actual dose that an individual internalized and relevance of that exposure cannot always be determined. The inherent problem of connecting these external exposures to internal biomarker measurements has always been a challenge and is further exacerbated with exposomic research. Newer approaches such as those used by Patel et al. may be helpful in this regard (55). Better models that integrate exposure data from multiple sources will be more useful for determining health impact. Additionally, surveys or questionnaires can be improved to assess current or past exposures in greater depth and accuracy to facilitate research findings.