In the past, exposures leading to diseases in humans were mainly assessed by questionnaires, which can be problematic as mentioned above.³⁴ It may be possible to at least partially counteract this challenge by looking for environmental samples with “historical relevance” indicating exposure, as shown in one of the next sections or just looking at human samples. Some governmental institutions collect and store human samples in biobanks over a period of time. Countries have been collecting and storing samples over 100 years without necessarily knowing the appropriate storage conditions or sample handling, especially for example, for methods in use now that were not available in the past.³⁵ The establishment of standards for storing different human samples began during the last 30 years, around the time when the term “biobank” was first used.³⁵ Using proper documentation of the samples makes dating much easier than it is for other sample types. However, for extremely old samples, such as mummies, other dating methods must be applied. Major issues of biobanks are sample degradation and missing long-time biobanks in many countries, as many efforts have just started collecting samples in the last few years.

Human samples contain a lot of information.⁵ However, it should be kept in mind that due to different metabolic processes, each sample matrix also reflects a different picture of the internal exposome, which in combination can give an overall picture (discussed further below). The most common samples, such as urine or blood, provide very transient signals for non-persistent chemicals, but can provide long-term information on environmental exposure for some persistent chemicals. Moreover, they have in the meantime well-established analytical protocols and standard materials available for analytical method development. Feces is a similarly non-invasive but relatively short-term sample type with a very complex matrix of increasing relevance given the attention to the microbiome; however, standard materials are still rare. Due to differences in diet, there is a high variability in stool samples; therefore, pooled samples have to be considered to perform individual or population studies. In the field of forensic toxicology, hair is used to provide valuable evidence on many aspects such as drug consumption that occurred, for example, up to weeks before.³⁶ Calafat et al.³⁷ show that baby teeth, amniotic fluid, and meconium can provide information on pre-natal exposures. Frye et al.³⁸ studied the connection of pre-natal metal exposure and autism looking at baby teeth. Spinal fluid and vitreous fluid, which is often used for post-mortem analyses, are two valuable but invasive sample types to mention; other sample types such as various tissues may also be available but are even more invasive. The main challenges with human samples are accessibility in sufficient quantity and the complex matrix that contains compounds such as endogenous metabolites in much higher concentrations than environmental pollutants.³⁹ Appropriate sample preparation and highly sensitive analytical techniques such as HR-MS are needed to analyze such samples. In Figure 1, the challenge of low concentrations of pollutants in biological samples is demonstrated compared with concentrations of drugs, endogenous, or food compounds. Another major issue is the comparability of different samples and measurements, as there is no standardized method to perform corrections for, for example, urinary dilution.^40,41 Moreover, there is no standard for reporting concentrations based on the different routes of elimination. These issues still need to be addressed in the context of exposomics, even before historical studies are performed.

Rappaport suggested to use a top-down approach in exposomic studies defining exposures as “biologically active chemicals” in the internal environment of a human organism.⁵ The original exposure, as well as its fingerprints in the form of metabolites and even detectable biomarkers are very valuable traces. Thus, even years after a smoker has stopped smoking, this past activity can be identified by certain changes in the human organism.^42,43 Unfortunately, this is not the case for all exposures. While metabolites can act as potential indicators for a certain disease or exposure (ie, functioning as biomarkers), they are not always new or sufficiently unique compounds. In many cases, exposure biomarkers may refer to elevated concentration of some compounds in the part of a population exposed to certain chemicals compared with lower levels in unexposed organisms as shown by Xu et al.⁴⁴ Not all exposures may have sufficiently specific biomarkers or associated changes in biochemical signals. However, the accessibility of “historical” human samples is a problem, as mentioned above: There are not always cohort studies or biobanks present that reach back to times of initial exposure; for instance, Luxembourg started to collect samples in 2009.⁴⁵ Moreover, not all types of exposures can be detected over a long time period in humans as for smokers, as metabolism is a very dynamic process such that most levels decrease over time, and metabolites are further transformed or eliminated.

Today, exposure assessment can be done efficiently by using wearables as shown by Hammel et al.⁴⁶ They can either function as sensors for environmental data or as monitoring devices for health data of the carrier.²⁹ The human organism is a complex sample itself, as it is influenced by many factors besides environmental pollutants; the biological response is a measure that can indicate such influences. Other factors as lifestyle, social factors, or other variables in the surrounding ecosystem play a major role, shown by example of cardiovascular risk factors as obesity or hypertension.^47,48

Other organisms may carry useful information about the present health or pollution state of the (aquatic) environment they are living in, for example, as demonstrated in mussels.⁴⁹ However, not much is known about backdating contaminants found in other organisms to the time of exposure, while not all species live sufficiently long. Some cetacean species live for several hundred years and have been shown to be exposed to environmental pollution from persistent chemicals many times over via biomagnification.⁵⁰ However, such studies mainly reflect the status and pollution levels of marine organisms.

When looking for representative plant species, trees stand out as promising specimen when it comes to finding historically relevant samples. Dendrochronologists can classify the exact age of trees via rings very accurately. It could be postulated that tree rings may yield data on air or soil pollution in chronological sections. Some studies deal exactly with this assumption, for example, Perone et al.⁵¹ showed in 2018 the temporal and spatial variability of air pollution from a wide range of sources in oak tree rings looking at tree cores. However, a study by the University of Göttingen in 2006 shows how problematic this assumption is.⁵² The enrichment of heavy metals in soils and the accompanying rising acidity increase the cation take-up of plants. Other factors such as growth rate and redistribution of elements determine element concentrations in each ring as well,⁵² confounding the interpretation. Some studies are also investigating the use of tree needles or leaves for biomonitoring of environmental pollution.^53,54 With the help of botanical collections even spatial and temporal trends can be found.

For environmental samples, so-called environmental specimen banks (ESBs) were established in many countries during the last 40 years.⁵⁵ In combination with this, there are digitally archived sample measurements available on repositories for HR-MS data such as Global Natural Products Social Molecular Networking (GNPS)⁵⁶ and NORMAN Digital Sample Freezing Platform (DSFP)⁵⁷ allowing retrospective analysis for recent years.⁵⁸ However, this will help future research, whereas historical environmental samples are most of the time not accessible any more.

The three most frequently examined sample types are soil/sediment, water, and air. Figure 2 shows the connection between these sample types and different contamination sources.

For all sample types, there are limitations: For water and sediment analysis, suitable water bodies are required. To look at human samples of the past, cohort studies must exist with suitable biobanks for the respective country or region. Other samples may not be sufficiently representative to monitor industrial pollution over a time period. Industrial pollution in soil, water, and air originating from past industries could be the reason for many diseases present nowadays. However, to connect those pollution events in the past to disease outcomes years later is a difficult task. Most of the time a combination of many factors plays a role in disease development, not just a handful of chemical compounds.²³ Exposomics research focuses on making these kinds of connections, which is a very challenging task that, in the context of historical exposomics, may require making the best of available information. Today, with the data and samples available presently, historical exposomics will involve estimating a reasonable exposure assessment for the past. For the future, proactively organized sampling campaigns could be used to gather and store all kinds of possible sample types at an early stage, to enable and simplify retrospective analyses. This type of sample retention should be in the interest of every population group, because the actual environmental cause of many diseases is often detected far too late and can then no longer be traced. If one also looks at the individual case, work-related illnesses, for example, in the military, can be traced back through previous exposure. In case one sample type is not present in a specific area, one can switch to other specimen indicating exposure, however, the comparability of different sample types is limited. Some pollutants only accumulate in certain media, like, for example, sediment, water, or fish (see Figure 4).

Metals are typical inorganic industrial contaminants. Their toxicity depends among other things on their total concentration as well as from the speciation of elements in the system.¹⁰¹ Even small amounts can bioaccumulate and have an influence on health. Many neurotoxins are metals such as aluminium (Al), arsenic (As), or mercury (Hg).¹⁰² Lead (Pb) contamination is a major issue especially in developing countries as its use is not regulated there.¹⁰³ It is persistent, widely used (eg, paints and cars) and therefore accumulates in the environment quickly, causing serious hazards all over the world. Organometallics have a broad range of applications in plastic manufacturing or as an additive to petrol in the past.¹⁰⁴ Tributyltin (a biocide) and methyl mercury (MeHg, formed by microbes or as a byproduct in industry) are just two high profile organometallic environmental contaminants to be mentioned. Organometallics possess a very high toxicity (eg, Minamata disease caused by MeHg¹⁰⁵), which is problematic as they are detectable in a variety of environmental samples through past or present use. An inductively coupled plasma (ICP) can be very useful as an ion source when analyzing metals via mass spectrometry (MS) and coupled to liquid chromatography (LC), even organometallics can be analyzed.¹⁰⁶

A larger number of organic contaminants are only relatively recently coming into focus, the so-called emerging pollutants: Pesticides like chlorpyrifos, per- and polyfluoroalkyl substances (PFAS), surfactants, pharmaceuticals or persistent, mobile, and toxic (PMT) substances in general, just to mention some groups. However, when investigating historical contamination, it is often difficult to determine the original concentration of some organic compounds and many might not be detectable any more. Other compounds, termed POPs accumulate in the environment over decades, such as the pesticide and insecticide dichloro-diphenyl-trichloroethane (DDT), which was banned in many countries in the 1970s.¹⁰⁷ However, regulation and therefore replacement of those chemicals often led to new emerging pollutants accumulating in nature,¹⁰⁷ with different transformation products that are not yet monitored (so-called regrettable substitution). Organic micropollutants at trace levels (µg/L to ng/L) have been released via anthropogenic activities to the environment over centuries and new substances are being discovered all the time. In a comprehensive annual 2020 review PFAS, replacement flame retardants, iodinated and nitrogenous dibutyl phthalates (DBPs), and antibiotic resistance genes (ARGs) were highlighted as groups of concern.¹⁰⁸ Many of the above mentioned analytes are associated with neurodegenerative diseases such as Alzheimer and Parkinson disease or certain cancers.¹⁰⁹ These connections were only found due to new technical improvements and methods.

The choice of the right method for each class of analytes is crucial to find contaminants even at trace levels. After sample preparation, chromatographic methods such as LC or gas chromatography (GC) are usually used to separate compounds of interest. MS is the detection tool of choice in most laboratories. Using different ion sources and mass analyzer modules such as Orbitraps can increase the sensitivity many times over. Before the actual analysis, the acquisition type must be determined. A distinction is made between targeted and non-targeted (NT) analyses.⁹³ Targeted analyses focus on a limited set of substances to be detected, where the reference standards are available in house in advance for method development. Some instruments are generally only used for targeted analysis (eg, triple quadrupole instruments), while others can offer both targeted and NT acquisition methods. For more details on analytical methods, several recent reviews, overviews, and comparisons exist.^110–113

Exposure can be calculated based on concentration values. However, for example, for biological samples correction methods to report concentrations are not (yet) harmonized and the values are therefore often not comparable.⁴⁰ For difficult matrices such as wastewater or feces, correction for matrix effects is essential to obtain reliable results.¹¹⁴ Moreover, an inter-batch correction compensating for varying signal intensities in a study is needed to compare measurements. Using signal intensities to quantify compounds measured with HRMS is highly problematic, since each compound ionizes differently (ionization efficiency can vary by up to six orders of magnitude)^115–117 and thus intensities are not directly comparable. Semi-quantification approaches such as structural similarity, parent—transformation product proximity, close eluting, ionization efficiency, or combined approaches can be used instead.¹¹⁸ Another major problem lies in the comparability of pollutant concentrations found in different media, as some pollutants only accumulate in specific matrices⁹⁵ (see Figure 4) and concentrations in organisms strongly depend on different metabolic processes. For exposomics studies, it is necessary to look at all types of pollutants and samples as, for example, viruses or bacteria in wastewater mirror the health status of a community⁸⁶ and consumer products in sediments⁹² reveal information about the lifestyle, each reflecting different sides of the exposome.

Using targeted methods with reference standards to quantify compounds does not necessarily cover the full range of substances in a sample, as important transformation products or pollutants that are not yet monitored may be omitted. However, HRMS is not required to perform targeted analysis. Targeted analysis on lower resolution instruments can be both cost and time efficient for routine analyses. Routine monitoring of, for example, certain rivers in targeted mode is needed to control if regulation values are met.¹¹⁹ While NT analysis covers more compounds and is more conducive to retrospective screening, is also not yet sufficiently harmonized and/or standardized for routine applications. For both historical exposomics and exposomics in general, precise definitions on how to measure each sample (number of repeats, choice of internal standards, or column, etc.) are necessary. Taking a critical look at the variety of existing methods in exposomics, there is some way to go before harmonization is sufficient for current studies, let alone for implementation into past studies. Some efforts at standardization of NTS are underway, which may pave the way for future harmonization efforts.¹²⁰

For NT MS data (hereafter NT-data), there are different data analysis options to consider: Targeted, suspect and NT screening. For targeted screening of NT-data, reference standards are required and matching MS data and retention time, preferably along with fragmentation (MS/MS) data are needed for identification of a compound.¹¹⁵ Suspect screening is the next potential step: A suspect list of several compounds, for example, pharmaceuticals is used and a search is made for matching MS and—if a library is used—MS/MS. However, if this list becomes too large one easily ends up in a NT approach, where peak-picking is performed, followed by identification efforts.¹¹⁵

There are many compound databases present for use in exposomics, which can be combined with spectral libraries that include spectra and thus fragmentation information for each compound for increased identification confidence.¹²¹ The largest compound databases now contain over 100 million entries, including CAS (184 million),¹²² PubChem (111 million),¹²³ and ChemSpider (114 million).¹²⁴ One example of a medium sized compound database that is often used in NT-HR-MS screening approaches, particularly in metabolomics is the Human Metabolome Database (HMDB), with HMDB4.0 containing 115 398 metabolite entries linked to 5702 protein sequences.^23,125 Major sources for suspect lists include the CompTox Chemical Dashboard with >300 lists and the NORMAN Suspect List Exchange with >80 lists, with individual lists containing 10 s up to >100 000 chemicals.^126–128 Using information of existing databases to generate new exposomic resources can be a useful approach to limit the number of chemicals considered in exposomics studies. The Blood Exposome Database was constructed using text mining and information from various databases, resulting in approx. 65 000 entries.¹²⁹ The Exposome Explorer, on the other hand, is a much smaller database of approx. 1000 entries.¹³⁰ Health-related databases using, for example, cohort studies or exposome databases like the Toxin-Toxin-Target Database (T3DB)¹³¹ can help to find a connection between exposures and health or specific phenotypes. As a database, T3DB is unique in that it shows mechanisms of toxicity as well as target proteins for each toxin, thus linking toxins (3678) and toxin targets (2073).¹³¹ PubChem are integrating many resources and presenting the interlinking of gene, protein, enzyme, disease, and chemical information as knowledge panels.¹³² Building a suspect list on, for example, industrial pollutants can be done by patent search of the PubChem¹²³ database and reducing the overlaps between different fields. The choice of the database always depends on the study question and often databases are too big. For some databases such as PubChem,¹²³ subsets exist to limit the number of compounds, for example, PubChemLite for Exposomics¹³³ contains the most relevant and annotated subset of chemicals in PubChem for exposomics. Besides compound databases, there are spectral libraries containing either experimentally or in silico predicted spectra of different compounds. Examples of databases containing compound and spectral information are GNPS,⁵⁶ MassBank of North America (MoNA),¹³⁴ National Institute of Standards and Technology (NIST),¹³⁵ METLIN,¹³⁶ and MassBank^137,138 with NIST 20 having for example 1.3 million tandem spectra compared with MoNA with 200 000 spectral records. More detailed numbers can be found in other articles on the topic.¹³⁹

A general issue lies within the use of databases: Using large databases yields many candidates per mass, providing many new ideas for possible chemicals, or leaving users juggling interpretations of various scoring terms. However, smaller databases containing just substances related to, for example, a disease or industrial use bear the risk of containing just “old knowledge” and not revealing any new knowledge. Thus, there is still a lot of work remaining until complete and comparable exposomics research is feasible, especially since harmonization and standardization is required in terms of terminology, methods, and reporting.

In untargeted analysis or suspect screening, there is the challenge of peak picking or feature detection, followed by annotation efforts to decipher the identity of the chemicals causing the features. A typical feature count can be of the order of tens of thousands of features per sample using NT-HR-MS.¹⁴⁰ There are several tools enabling (partially) automated data analysis, for example patRoon,¹⁴¹ XCMS,¹⁴² or MS-DIAL¹⁴³ (see Figure 5) that can be used to look for specific masses or compounds present in a sample. In 2019, Wang et al.¹⁴⁰ presented PAVE, a peak annotation and verification engine for metabolomics, which includes the “cleaning” of the data and results in the matching metabolite formula. However, this approach requires stable isotope labeling, which is not feasible for most specimen in exposomics.

One speaks intentionally of peak annotation instead of identification as a full identification of a chemical can only be achieved using reference standards (ie, confirmation with target compounds).⁹³ However, as many standards are difficult to obtain, feature annotation using different computational tools is an alternative approach to tentatively identify chemicals of interest for further confirmation efforts. Feature annotation and compound identification are based on different parameters: The exact mass of the compound paired with its fragmentation pattern can be compared, for example, to experimental or in silico spectra using open source software such as MS-DIAL¹⁴³ or MetFrag.¹⁴⁴ In addition, retention time can improve identification, depending strongly on the method and instrument used.¹⁴⁵ Blaženović et al. used for their analysis of urinary metabolites a combination of several computational tools as CSI: FingerID¹⁴⁶ or NIST hybrid search¹⁴⁷ in order to annotate all metabolites found.¹³⁹ There are many other ways to annotate features; the software approaches provided by vendors of MS devices are also a good option for many, but are not covered in detail here. Figure 5 shows various ways of analyzing NT MS data resulting in the different identification levels.¹²¹

Usually, a statistical analysis follows after annotation (sometimes even before), including uni- and multivariate analysis as well as a validation of the study design and the interpretation of the results. The statistical evaluation methods will not be further elaborated here.¹⁴⁸ Statistics can be used to understand the individual, community, and ecological exposome changes over time and the resulting health outcomes by interconnecting different data types and analyzing exposure values. This is required in order to understand environmental influences and their impact on health better (eg, for occupational diseases) and to act according to precautionary principles with regards to certain prevention measures (regulations, etc.). Typical strategies in exposomics studies include risk-based prioritization of chemicals and estimating the environmental disease burden.¹⁴⁹ Models for exposure risk and hazard assessment can then be applied in environmental policies and regulations.

There are computational tools in development to enter exposure data and analyze it using standard methods from, for example, Bioconductor.¹⁵⁰ The R-package “rexposome” can be used to connect exposures to phenotypes in exposome association studies.¹⁵¹ The input data in such tools can be acquired by many disciplines, not just chemically using HR-MS. HExpMetDB was developed as a risk-prioritized human exposome database containing physiochemical properties and risk prediction with a graphical user interface (GUI) which enables searching.¹⁵² Finding the connection to health is quite challenging. Health-related databases—as the ones mentioned above—can assist in finding this connection. Health records can be of great value when looking at past events. However, it is not just about doing an epidemiological study looking at several pollutants in connection to a health risk.

The interconnection of exposures in chemical networks as well as other factors influencing the health of an organism have to be analyzed.²³ Therefore, environment-wide association studies and even exposome-wide association studies are appropriate ways of finding the connection between exposure and health.^23,153 Metabolome wide association studies find connections between metabolic profiles and disease risk, look for biomarkers of exposure, and even predict future disease onset. However, for all those studies, it is challenging to find relationships between thousands of molecular markers and disease phenotypes with minimal false positive associations.¹⁵⁴ Analytical techniques such as HR-MS or nuclear magnetic resonance enable metabolic profiling and exposure assessment. This makes association studies, metabolic pathway enrichment as well as looking at molecular networks possible.¹⁵⁵ Machine learning can help recognize patterns and make predictions thereafter. However, the limitations of all those approaches have to be considered: Dealing with thousands of features per sample, annotation and identification (with a certain confidence) become difficult and there is still a lack of automation. Moreover, the diversity of chemicals and chemical mixtures has to be taken into account with many unknown variables remaining.¹⁵⁶ Today, it is no longer a problem to measure the samples with sufficient sensitivity in a short time, but rather to draw the right conclusions from the results. Another issue is the terminology that differs between the research fields and the urgent need for harmonization. Association studies require strong international collaborations and high level networking, which is nearly impossible without establishing common terminology.¹⁵⁶

Spatial and geographic data can be used in many ways for exposomics purposes. Geographical information systems (GIS) can connect different types of information that seem completely unrelated.²⁹ For exposomics it can be helpful to look at various sample types as presented above, at literature and health records to derive facts about historical exposure from those sources and link them geographically. Historical maps or aerial photographs often provide a good source of information on possible contaminated areas to establish connections to industrial sites, landfills, main traffic routes, and bigger cities as these may be more likely to have high levels of pollutants.²⁹ GIS helps in combining these different information sources by overlaying maps or aerial photographs, integrating data of environmental exposures and health-related data, and analyzing changes over time.²⁹ Distances between the source of exposure, for example, a closed landfill and affected people can be monitored,¹⁵⁷ as well as mobility of people and factors such as the density of grocery stores offering healthy options.¹⁵⁸ GIS can help in risk assessment and future planning as well as with environmental models.¹⁵⁹ Presenting information on environmental issues in a spatial and graphical way makes analysis easier and enables planning for future needs.