Trends in human risk assessment of pharmaceuticals
- First Online:
- Cite this article as:
- Dorne, J.L.C.M., Ragas, A., Frampton, G. et al. Anal Bioanal Chem (2007) 387: 1167. doi:10.1007/s00216-006-0961-9
- 192 Views
Pharmaceutical products used in human and veterinary medicine are a class of great importance in our modern society. The scientific community has recently recognized that after elimination from the body of humans and domestic animals, active ingredients are found in treated and untreated sewage effluents, surface water, groundwater, and drinking water . Many therapeutic classes are commonly found, for example anti-inflammatory drugs, cholesterol-lowering drugs (e.g. statins), antidepressants, anticonvulsants, synthetic steroids (e.g. the contraceptive pill), antineoplastics, beta-blockers, bronchodilators, lipid regulators, hypnotics, antibiotics, antiseptics, X-ray contrast agents, and caffeine. Such contamination should be assessed in relation to possible environmental and human risks, because pharmaceutical products are biologically active, for example synthetic hormones which cause endocrine disruption in fish at very low levels of exposure (ng L−1) . Recent reviews describing detrimental effects of pharmaceutical compounds on the ecosystem are available, and also presented in this issue [3, 10, 13]. Advanced technologies using granular activated carbon, membrane technology, ozonation, and ultraviolet radiation have been used with relative success to remove pharmaceutical and environmental contaminants from water destined for human consumption. Several pharmaceutical products, for example anti-epileptics (carbamazepine, primidone), non-steroidal anti-inflammatory drugs (diclofenac, ibuprofen, ketoprofen, indomethacin), and lipid regulators (clofibric acid, gemfibrozil) are known to resist such treatment, because of their high solubility and/or poor degradability in water. Hence, concentrations of pharmaceutical products in drinking water have been shown to reach ng L−1 levels, and even μg L−2 levels for compounds such as diclofenac, the pharmaceutical compound present at the highest concentration in the environment . Risk assessments for potential adverse effects in humans at these concentrations are scarce particularly for chronic exposure and sub-therapeutic concentrations. Such assessments should become a scientific priority for the “International Decade for Action: Water for Life” launched in 2005 and the Millennium Development Goals (MDGS), the objective of which is to double the amount of the world’s population with sustainable access to safe drinking water and sanitation by 2015. The UNICEF and WHO have estimated that over a billion and 2.6 billion people are currently deprived of safe water supplies and adequate sanitation, respectively .
Trends in the risk assessment of pharmaceutical products for human health are addressed here for non-cancer and cancer outcomes, including the threshold and safety factor approach, mixtures of pharmaceutical products, and use of emerging technology, for example toxicogenomics, quantitative structure–activity relationships, and probabilistic methods.
Human risk assessment of pharmaceutical products—trends and future refinements
The objective of human risk assessment is to protect susceptible individuals of the human population from potential chemical harm by derivation of safe levels of exposure “without appreciable health risk”, for example the acceptable or tolerable daily intake (ADI and TDI) (WHO) or the reference dose (RfD) (US-EPA). For non-genotoxic carcinogens present in food and water contaminants, these have been set using thresholds (below which no toxicity is predicted), expressed in mg kg−1 diet or volume of water per day, to relate them to human oral exposure. The surrogate for the threshold, for example the lowest and/or no observed adverse effect level (LOEL, NOEL) or the benchmark dose (BMD) is determined from chronic or subchronic animal studies and then divided by an uncertainty factor of 100 (interspecies differences tenfold and human variability tenfold) to derive the ADI or RfD. For genotoxic carcinogens, dose–response relationships from experimental animal data are combined with low dose extrapolation to relate a human health risk to an estimated exposure or an estimated exposure to a human health risk ([9, 11].
Guideline values and Ambient Water Quality Criteria thresholds have also been developed by the US-EPA and WHO for pharmaceutical products, based on therapeutic effects and adverse effects, i.e. probable idiosyncratic and/or hypersensitivity reactions. In the WHO approach the guideline value is derived using daily water consumption, the fraction of the tolerable daily intake allocated to water consumption, and the body weight of the subjects . Uncertainty factors (30) are applied to take into account human variability (tenfold) and extrapolation from LOEL to the NOEL derived from animal data (threefold).
These data are based on thirty pharmaceuticals but do not cover other classes, for example pharmaceuticals used in veterinary medicine.
Some pharmaceutical products have been designed for specific subgroups of the human population. Therapeutic doses and toxic doses may vary widely between subgroups and these should also be assessed for the most susceptible subgroups, for example neonates .
These human health risk assessments have been conducted for given scenarios of exposure in water sources yet levels of pharmaceuticals may fluctuate in different regions of the globe and/or bioaccumulate, depending on human activity and therapeutic use of pharmaceutical compounds. Hence, regular biomonitoring of concentrations of pharmaceutical products in water sources is also a major priority.
Pharmaceuticals are always present as mixtures, an issue which still greatly troubles regulatory agencies ; a major challenge to scientists is to evaluate whether these mixtures in surface and ground water could constitute a health risk to humans now or in the future.
Finally, another important aspect is the scientific validity of the tenfold uncertainty factor allowing for human variability and used to derive safe levels of exposure for food and water contaminants for humans. Over the last ten years, regulators and scientists have subdivided the human variability factor (10) into toxicokinetics (TK, (100.5, 3.16), elimination of compounds and toxicodynamics (TD), and differences in the expression of toxicity (100.5, 3.16) . Depending on the availability of chemical-specific data, such factors could be replaced with chemical specific adjustment factors (CSAFs) usually developed from a physiologically based TK-TD model (PB-TK-TD). The human body uses different pathways to eliminate compounds and a recent approach has derived pathway-related uncertainty factors to replace the TK uncertainty factor for each route and for different percentiles (95–99%) of human subgroups (genetic polymorphism, disease, age and ethnicity). Such an approach enables incorporation of in vivo human metabolism and TK data in risk assessment and provides flexible options between uncertainty factors and CSAFs. This approach has been based on meta-analyses of human studies describing the TK of pharmaceuticals metabolised primarily (>60% dose) by major human phase-I (mostly CYP) metabolism, phase-II hepatic metabolism, and renal excretion. The phase-I metabolic routes include the major human CYP isoforms (CYP1A2, CYP2E1, CYP3A4, CYP2C9, CYP2C19, CYP2D6), alcohol dehydrogenase (ADH), and major hydrolysis metabolic routes. Phase-II metabolism includes N-acetylation (NAT-2), glucuronidation, glycine and sulfate conjugation, and renal excretion . Overall, the TK uncertainty factor would not cover neonates, the elderly for most elimination routes, and any human subgroup for compounds metabolised by polymorphic CYP (for example as CYP2C9, CYP2C19, CYP2D6, and NAT-2).
These conclusions raise questions about the safety of pharmaceuticals in water, as individuals vary in their susceptibility. This is of particular relevance to compounds metabolised by polymorphic CYP enzymes for which elimination differences are very wide with over 3–20-fold differences between extensive metabolisers (EMs) and poor metabolisers (PMs). Extra uncertainty factors would have a large impact on the setting of safe levels and it is important that future approaches take these quantitative differences into account. For polymorphic metabolism it can be assumed that the proximate toxicant is the parent compound and PMs would be the susceptible subgroup, i.e. a decrease in clearance would increase adverse effect risk. The reverse situation is also commonly seen, however, i.e. metabolic activation to a toxic species so that EMs would be the susceptible subgroup .
Metabolism and pharmacodynamics for several pharmaceutical products of environmental relevance
Therapeutic/ chemical class
PD mode of action
Pharmaceutical product substrates
Inhibition of phosphodiesterase
Nicotinic receptor agonist
CYP2C9, glucuronidation, renal excretion
Inhibition of cyclooxygenase
Ibuprofen, indomethacin, diclofenac
CYP2C9, renal excretion
Reducing glucose levels
Inhibition of blood coagulation factors
CYP2C19, CYP3A4, renal excretion
Inhibition of plasmodium growth
CYP2D6, CYP2C19, CYP3A4, CYP1A2
Inhibition of serotonin reuptake
Fluoxetine, paroxetine, citalopram
Inhibition of H+-ATPase pump
Inhibition of bacterial growth
Inhibition of bacterial growth
Reducing cholesterol levels
Lovastatin, simvastatin, fluvastatin
Human risk assessment of pharmaceutical mixtures
CYP-based interactions and their toxicokinetic and toxicodynamic consequences
Mechanism of TK interaction
Increased risk of adverse effects
Increase in heart rate
Reduce ADP-induced platelet aggregation
CYP2D6, CYP1A2, glucuronidation
Loss of dysrhythmia control in heart
St John’s wort
Risk of rejection of organ transplants
Reduce effect but not tested
St John’s wort
Increase bleeding time
These examples are based on therapeutic concentrations (mg day−1) that would be higher than environmental exposure to pharmaceutical products (ng or μg day−1). The effects of chronic exposure to low or sub-therapeutic doses of pharmaceutical products in the presence of potent competitive inhibitors (for example paroxetine for CYP2D6, protease inhibitors (ritonavir, indinavir) and antifungal drugs (ketoconazole, itraconazole) for CYP3A4) or inducers has not yet been assessed, however. A recent study has shown that current levels of exposure to organophosphate pesticides (<10 μmol L−1) inhibit metabolism of the antidepressant imipramine by human recombinant enzymes and in liver microsomes . This raises important issues for the risk assessment of mixtures and the potential for interaction between pharmaceutical products and other environmental contaminants. The human health relevance of these observations must be assessed more thoroughly with a wide range of pharmaceutical products, pesticides, and other environmental contaminants as potential inhibitors or inducers of drug-metabolising enzymes.
One of the objectives of the 6th framework European project NOMIRACLE (http://viso.jrc.it/nomiracle/) is to improve the scientific basis of uncertainty factors with particular reference to chemical mixtures. Recent analysis of TK interactions between probe substrates of polymorphic CYPs (CYP2C9, CYP2C19 and CYP2D6) and known inhibitors or inducers, has shown that the current TK uncertainty factor (3.16) would not cater for such interactions if the parent compound or its metabolite(s) were the proximate toxicant. Although these results are based on therapeutic concentrations rather than the low concentrations of pharmaceutical products in drinking water, genetic variability and potential interactions (including potency of the inhibitor or inducer) should be taken into account in human risk assessment. This may be of particular relevance to anti-cancer drugs and endocrine disrupters present in the environment as potential carcinogens, and their effect(s) may be modified during chemical interactions . Such information can be obtained routinely in the laboratory by use of recombinant technology and toxicokinetic assays .
Future approaches: toxicogenomics, QSAR 282and probabilistic approaches
The emergence of the field of systems biology, seeking to integrate different levels of biological information to understand how biological systems function, has introduced to toxicology a set of experimental methods and tools with potential application for refining and improving approaches to risk assessment. A recent workshop on toxicogenomic methods involving European and US experts, regulators, and principal validation bodies has addressed biological validation of such methods in the regulatory arena. Toxicogenomics and its cousin disciplines (proteomics, transcriptomics, metabolomics, and metabonomics) can help define mechanisms and identify biomarkers in combination with gene expression signatures to predict toxicity. Although such data are, currently, often derived from in vitro assays and have limited regulatory applications, they would still enable prescreening and complementary toxicological tests for single compounds and priority mixtures, i.e. microarrays indicating changes in gene expression after chemical exposure (potentially defining mechanisms of action, and potential alternatives to refine, reduce, and replace animal use). Other long-term goals can also be foreseen, for example the use of human or animal in vitro or in vivo data derived from omic technology to derive adverse (or no adverse) effect levels or to determine dose–response relationships for quantitative risk assessments .
Combining toxicogenomics data with other toxicological information, for example quantitative structure–activity relationships, can also provide very useful information for scientists assessing pharmaceutical products. The objective of quantitative structure–activity relationships (QSARs) is to correlate the structure of chemicals with their activity by use of statistical tools. These in silico approaches can be of great value for predicting the activities of chemicals that have not been tested, including pharmaceutical products . QSAR models with good predictive power (>92%), using clinical trials and maximum recommended therapeutic doses (MRTD) for high and low-toxicity compounds, have been recently been developed for estimation of no observed effect levels (NOELs) of pharmaceutical products in humans. Such human data provide more specific estimates of toxic dose thresholds of chemicals in humans than do extrapolated animal data; e.g. there is poor correlation between MRTD in rodents and humans (R2 = 0.2005, n = 326) . QSAR is a very powerful tool for molecular modelling of enzyme–substrate interactions when investigating CYP metabolism, because quantitative information is available on the linear relationship between the strength of substrate binding and the hydrophobicity log P (where P is the octanol/water partition coefficient), on hydrogen bonding, and on aromatic π–π stacking . QSAR models have also been developed for CYP3A4, which metabolises 50% of all known pharmaceuticals, and known competitive inhibitors, and there is good agreement between these models, molecular models of the enzyme itself, and known mechanisms of inhibition .
Science-based risk assessment has benefited from use of probabilistic risk assessment techniques which can replace “single value deterministic uncertainty factors” by distributions and quantify the uncertainty of the assessment explicitly. The most popular technique is Monte Carlo modelling in which the deterministic input values of the assessment (i.e. human uncertainty factors) are replaced by input distributions and, subsequently, an output distribution of risk is produced by repetitive drawings from the input distributions . Such models have enabled good prediction of human variability in kinetics (and uncertainty factors) for compounds handled by multiple pathways of elimination with known in-vitro and in-vivo quantitative metabolic profiles in man [7, 8].
Monte Carlo simulation can also be combined with Bayesian statistics, enabling the combination of different sources of probabilistic information in one assessment. An example is the study by Roelofs et al.  in which an ecological NOEL is predicted on the basis of a combination of substance-specific toxicity data and non-substance-specific information from a toxicity database. These techniques can have disadvantages, however; e.g. different types of input variation are combined into one output distribution which is a mixture of the variance and the uncertainty of the output. One means of looking at both aspects is use of 2nd-order Monte Carlo simulation  so that uncertainty and variability are nested in two different simulations and can be combined in a final distribution. This type of analysis can produce quantitative results giving information about chemical risk, i.e. “10% probability (= uncertainty) that 5% of the population (interindividual variability) exceeds the safe concentration”.
Conclusion and future work
Assessment of risk to humans of environmental exposure to pharmaceutical products (mostly drinking water and fish) can be regarded as a new field of risk assessment and risk analysis characterised by its inherent multidisciplinary nature. The link between human and ecological risk assessment is clear, because the effect of chemical species in the environment may have direct or indirect effects on human health. Advances in toxicology have helped scientists define pathways of metabolism (toxicokinetics) and mechanisms of toxicity (toxicodynamics) for single pharmaceutical products and, increasingly, for mixtures. New tools, for example the “omic” sciences, QSAR modelling, and probabilistic models have great potential to assist scientists in quantifying biological effects of pharmaceutical products and identifying toxicity mechanisms at the level of populations, individuals, cells, and molecular targets. These results can then be used to refine uncertainty factors used to set safe human levels or, ideally, derive chemical-specific adjustment factors.
This manuscript is dedicated to the memory of Professor John Theobald (1946–2006).
Some of the authors (JLCMD, AMJR, DSS) are grateful to the European Commission under the NOMIRACLE project (number 003956) for funding this work. The opinions reflected in this review are the authors’ only.