key: cord-0001244-59i3jert
authors: Ashbolt, Nicholas J.; Amézquita, Alejandro; Backhaus, Thomas; Borriello, Peter; Brandt, Kristian K.; Collignon, Peter; Coors, Anja; Finley, Rita; Gaze, William H.; Heberer, Thomas; Lawrence, John R.; Larsson, D.G. Joakim; McEwen, Scott A.; Ryan, James J.; Schönfeld, Jens; Silley, Peter; Snape, Jason R.; Van den Eede, Christel; Topp, Edward
title: Human Health Risk Assessment (HHRA) for Environmental Development and Transfer of Antibiotic Resistance
date: 2013-07-09
journal: Environ Health Perspect
DOI: 10.1289/ehp.1206316
sha: aef400eac72623d01820a1b2ca11e49a81917416
doc_id: 1244
cord_uid: 59i3jert

Background: Only recently has the environment been clearly implicated in the risk of antibiotic resistance to clinical outcome, but to date there have been few documented approaches to formally assess these risks. Objective: We examined possible approaches and sought to identify research needs to enable human health risk assessments (HHRA) that focus on the role of the environment in the failure of antibiotic treatment caused by antibiotic-resistant pathogens. Methods: The authors participated in a workshop held 4–8 March 2012 in Québec, Canada, to define the scope and objectives of an environmental assessment of antibiotic-resistance risks to human health. We focused on key elements of environmental-resistance-development “hot spots,” exposure assessment (unrelated to food), and dose response to characterize risks that may improve antibiotic-resistance management options. Discussion: Various novel aspects to traditional risk assessments were identified to enable an assessment of environmental antibiotic resistance. These include a) accounting for an added selective pressure on the environmental resistome that, over time, allows for development of antibiotic-resistant bacteria (ARB); b) identifying and describing rates of horizontal gene transfer (HGT) in the relevant environmental “hot spot” compartments; and c) modifying traditional dose–response approaches to address doses of ARB for various health outcomes and pathways. Conclusions: We propose that environmental aspects of antibiotic-resistance development be included in the processes of any HHRA addressing ARB. Because of limited available data, a multicriteria decision analysis approach would be a useful way to undertake an HHRA of environmental antibiotic resistance that informs risk managers. Citation: Ashbolt NJ, Amézquita A, Backhaus T, Borriello P, Brandt KK, Collignon P, Coors A, Finley R, Gaze WH, Heberer T, Lawrence JR, Larsson DG, McEwen SA, Ryan JJ, Schönfeld J, Silley P, Snape JR, Van den Eede C, Topp E. 2013. Human health risk assessment (HHRA) for environmental development and transfer of antibiotic resistance. Environ Health Perspect 121:993–1001; http://dx.doi.org/10.1289/ehp.1206316

A workshop (Antimicrobial Resistance in the Environment: Assessing and Managing Effects of Anthropogenic Activities), held in March 2012 in Québec, Canada, focused on anti biotic resistance in the environment and approaches to assessing and managing effects of anthropogenic activities. The human health concern was identified as environmentally derived antibioticresistant bacteria (ARB) that may adversely affect human health (e.g., reduced efficacy in clinical anti biotic use, more serious or prolonged infection) either by direct exposure of patients to antibiotic resistant pathogen(s) or by exposure of patients to resistance determinants and subsequent horizontal gene transfer (HGT) to bacterial pathogen(s) on or within a human host, as conceptualized in Figure 1 . ARB hazards develop in the environment as a result of direct uptake of antibioticresistant genes (ARG) via various mechanisms (e.g., mobile genetic elements such as plasmids, integrons, gene cassettes, or transposons) and/or proliferate under environmental selection caused by anti biotics and coselecting agents such as biocides, toxic metals, and nanomaterial stressors (Qiu et al. 2012; Taylor et al. 2011) , or by gene mutations (Gillings and Stokes 2012) . Depending on the presence of recipient bacteria, these processes generate either environmental antibioticresistant bacteria (eARB) or pathogens with antibioticresistance (pARB) (Figure 1 ).

Human health risk assessment (HHRA) is the process used to estimate the nature and probability of adverse health effects in humans who may be exposed to hazards in contaminated environmental media, now or in the future [U.S. Environmental Protection Agency (EPA) 2012]. In this review we focus on how to apply HHRA to the risk of infec tions with pathogenic ARB because they are an increasing cause of morbidity and mor tality, particularly in developing regions Background: Only recently has the environment been clearly implicated in the risk of antibiotic resistance to clinical outcome, but to date there have been few documented approaches to formally assess these risks. oBjective: We examined possible approaches and sought to identify research needs to enable human health risk assessments (HHRA) that focus on the role of the environment in the failure of anti biotic treatment caused by antibiotic-resistant pathogens. Methods: The authors participated in a workshop held 4-8 March 2012 in Québec, Canada, to define the scope and objectives of an environmental assessment of antibiotic-resistance risks to human health. We focused on key elements of environmental-resistance-development "hot spots," exposure assessment (unrelated to food), and dose response to characterize risks that may improve antibiotic-resistance management options. discussion: Various novel aspects to traditional risk assessments were identified to enable an assessment of environmental antibiotic resistance. These include a) accounting for an added selective pressure on the environmental resistome that, over time, allows for development of antibioticresistant bacteria (ARB); b) identifying and describing rates of horizontal gene transfer (HGT) in the relevant environmental "hot spot" compartments; and c) modifying traditional dose-response approaches to address doses of ARB for various health outcomes and pathways. conclusions: We propose that environmental aspects of antibiotic-resistance development be included in the processes of any HHRA addressing ARB. Because of limited available data, a multicriteria decision analysis approach would be a useful way to undertake an HHRA of environmental antibiotic resistance that informs risk managers. citation: Ashbolt NJ, Amézquita A, Backhaus T, Borriello P, Brandt ). An antimicrobial resistant micro organism has the ability to mul tiply or persist in the presence of an increased level of an anti microbial agent compared with a susceptible counter part of the same species. For this review, we limited the resistant group of micro organisms to bacteria and therefore to anti biotic resistance, an area in which the term "antibiotic" is used synonymously with "antibacterial." It is important to understand the contribution that the environment has on the development of resistance in both human and animal pathogens because therapeutic resistant infections may lead to longer hos pitalization, longer treatment time, failure of treatment therapy, and the need for treatment with more toxic or costly antibiotics, as well as an increased likelihood of death.

A vast amount of work has been under taken to understand the contribution and roles played by hospital and community settings in the dissemination and maintenance of ARB infections in humans. A particular area of focus in terms of exposure in a community setting has been anti biotic use in livestock produc tion and the presence of eARB and pARB in food of animal origin. In 2011, the Codex Alimentarius Commission [established in 1963 by the Food and Agriculture Organization of the United Nations (FAO) and the World Health Organization (WHO) to harmonize international food standards, guidelines, and codes of practice to protect the health of con sumers and ensure fair trade practices in the food trade] released guidelines on processes and methodologies for applying risk analy sis methods to foodborne anti microbial resis tance related to the use of anti microbials in veterinary medicine and agriculture (Codex Alimentarius Commission 2011).

Other sources of anti biotics and other anti microbials in the environment are human sewage (Dolejska et al. 2011) , intensive ani mal husbandry, and waste from the manu facture of pharmaceuticals (Larsson et al. 2007 ). The environmental consequences from the use and release of anti biotics from various sources (Kümmerer 2009a (Kümmerer , 2009b and the HGT of antibioticresistance genes (ARG) between indigenous environmental and pathogenic bacteria and their resistance determinants (Börjesson et al. 2009; Chagas et al. 2011; Chen et al. 2011; Cummings et al. 2011; Forsberg et al. 2012; Gao et al. 2012; Qiu et al. 2012 ) has yet to be quanti fied, but is of global concern (Finley et al. 2013; WHO 2012a) . The genetic elements encoding for the ability of micro organisms to withstand the effects of an anti microbial agent are located either chromosomally or extra chromosomally and may be associated with mobile genetic elements such as plas mids, integrons, gene cassettes, or transpo sons, thereby enabling horizontal and vertical transmission from resistant to previously susceptible strains. From an HHRA point of view, the emergence of ARB in source and drinking water (De Boeck et al. 2012; Isozumi et al. 2012; Shi et al. 2013 ) further highlights the need to place these emerging environmental risks in perspective. Yet, assess ing the range of environmental contribu tions to anti biotic resistance may not only be complicated by lack of quantitative data but also by the need to coordinate efforts across different agencies that may have jurisdiction over environmental risks versus human and animal health.

A key consideration for ARB develop ment in the environment is that resistance genes can be present due to natural occur rence (D'Costa et al. 2011 ). Further, the use of anti microbials in crops, animals, and humans provides a continued entry of anti biotics to the environment, along with pos sible novel genes and ARB. A summary of the fate, transport, and persistence of antibiotics and resistance genes after land application of waste from food animals that received antibiotics or following outflow to surface water from sewage treatment has emphasized the need to better understand the environ mental mechanisms of genetic selection and gene acquisition as well as the dynamics of resistance genes (resistome) and their bacte rial hosts (CheeSanford et al. 2009; Crtryn 2013) . For example, the presence of anti biotic residues in water from pharma ceuti cal manufacturers in certain parts of the world (Fick et al. 2009 ), ponds receiving intensive animal wastes (Barkovskii et al. 2012) , aqua culture waters (Shah et al. 2012) , and sewage outfalls (Dolejska et al. 2011 ) are important sources, among others, leading to the pres ence of ARG in surface waters. In particu lar, the comparatively high concentrations of anti biotics found in the effluent of pharma ceuti cal production plants have been asso ciated with an increased presence of ARG in surface waters (Kristiansson et al. 2011; Li et al. 2009 Li et al. , 2010 . Most recently, 100% sequence identity of ARG from a diverse set of clinical pathogens and common soil bacte ria (Forsberg et al. 2012 ) has highlighted the potential for environ mental HGT between eARB and pARB.

Despite these concerns, few risk assess ments have evaluated the combined impacts of anti biotics, ARG, and ARB in the environ ment on human and animal health (Keen and Montforts 2012) . Recent epidemiological stud ies have included evaluation of ARB in drink ing water and the susceptibility of commensal Escherichia coli in household members. For example, Coleman et al. (2012) reported that water, along with other factors not directly related to the local environment, accounted for the presence of resistant E. coli in humans. In many studies, native bacteria in drinking water systems have been shown to accumulate ARG (VazMoreira et al. 2011) .

In addition to addressing environmental risks arising from the development of anti biotic resistance, we should also consider the or Development and enrichment of pARB low probability but high impact "onetime event" type of risk. This exceedingly rare event that results in the transfer of a novel (to clinically important bacteria) resistance gene from a harmless environmental bacterium to a pathogen need happen only once if a human is the recipient of the novel pARB. Unlike the emergence of SARS (severe acute respira tory syndrome) and similar viruses where, in hindsight, the risk factors are now well under stood (Swift et al. 2007 ), the conditions for a "onetime event" could occur in a range of "normal" habitats. Once developed, the resis tant bacterium/gene has a possibility to spread between humans around the world [such as seen with the spread of NDM1 (New Delhi metallobetalactamase1) resistance (Wilson and Chen 2012) ], promoted by our use of anti biotics. Although it seems very difficult to quantify the probability for such a rare event (including assessing the probability for where it will happen and when), there is consider able value in trying to identify the risk factors (such as pointing out critical environments for HGT to occur, or identifying pharmaceutical exposure levels that could cause selection pres sures and hence increase the abundance of a given gene). After such a critical HGT event, we may then move into a more quantitative kind of HHRA. The overall goal of the workshop (Anti microbial Resistance in the Environment: Assessing and Managing Effects of Anthropogenic Activities) was to identify the significance of ARB within the environment and to map out some of the complexities involved in order to identify research gaps and provide statements on the level of scientific understanding of various ARB issues. A broad range of international delegates, including aca demics, government regulators, industry mem bers, and clinicians, discussed various issues. The focus of this review arose from discussions of improving our understanding of human health risks-in addition to epidemiological studies-by developing HHRAs to explore potential risks and inform risk manage ment. Because the end goal of an assessment depends on the context (e.g., research, regulation), we provide a generic approach to under taking an HHRA of environmental ARB that can be adapted to the users' interest (conceptualized in Figure 1 ). Given the many uncertainties, we also highlight identified research gaps.

Understanding other on going relevant inter national activities and the types of anti biotics used provide good starting points to aid in framing a risk assessment of ARB. The Codex Alimentarius Commission (2011) described eight principles that are specific to risk analysis for foodborne anti microbial resistance, several of which are generally applicable to a HHRA of environ mental ARB. Examples include the recommendations of the Joint FAO/WHO/ OIE Expert Meeting on Critically Important Antimicrobials (Food and Agriculture Organization of the United Nations/World Health Organization/World Organisation for Animal Health 2008) and the WHO Advisory Group on Integrated Surveillance of Antimicrobial Resistance (WHO 2012b), which provided information for setting the priority anti biotics for a human risk assess ment. It should be noted that there are sig nificant national and regional differences in anti biotic use, resistance patterns, and human exposure pathways.

In general, risk assessments are framed by identifying risks and management goals, so the assessment informs the need for possible management options and enables evaluation of management success. The consensus of workshop participants was that manage ment could best be applied at points of anti biotic manufacturing and use, agricultural operations including aquaculture, and wastewater treat ment plants (Pruden et al. 2013) . Assessing the relative impact of managing any particular part of a system is hampered by the lack of knowledge on the relative importance of each part of the system for the overall risk. That is, as recently stated by the WHO (2013), "AMR is a complex problem driven by many inter connected factors so single, isolated interventions have little impact and coordi nated actions are required." Hence, a start ing point for an assessment of environmental anti bioticresistance risks intended to aid risk management is a theo retical risk assessment pathway based on a) local surveillance data on the occurrence and types of anti biotics used in human medi cine, crop production, animal husbandry, and companion animals; b) infor mation on ARG and ARB in the various environmental compartments (in particular, soil and aquatic systems including drinking water); and c) related disease information. This assessment should be amended by discussion with the relevant stakeholders, which requires extensive risk communication and could form part of the multi criteria decision analysis (MCDA) approach discussed in detail below. As a result of the workshop, Pruden et al. (2013) also advocate coupling environ mental manage ment and mitigation plans with tar geted surveillance and monitoring efforts in order to judge the relative impact and success of the interventions.

To undertake a useful human health risk assessment, some details require quantitative measures. Thus, the key issue is how experi mental and modeling approaches can be used to derive estimates. Furthermore, haz ard concentration, time, and environ mental compartmentdependent aspects should also be taken into account. First, the current understanding is that for nonmutation derived antibiotic resistance to develop in environmental bacteria (including pathogens that may actively grow outside of hosts) to develop into eARB/pARB ( Figure 1 , pro cesses 1 and 2), a selective pressure (i.e., pres ence of anti biotics or antibioticresistance determinants) must be maintained over time in the presence of ARG; for existing pARB released into the environment, sur vival in environmental media is the critical factor. However, the exact mechanisms and quantitative relationships between selective pressures and ARB development have yet to be elucidated, and they may be different depending on the anti biotic, bacterial spe cies, and resistance mechanisms involved. In cases where selective pressure is removed, the abundance of antibioticresistance ARB may be reduced, but not to extinction. Hughes 2010, 2011; Cottell et al. 2012) . Even a small number of ARB at the com munity level represents a reservoir of ARG for horizontal transfer once pressure is reap plied. Because it seems inevitable that ARB will eventually develop against any anti biotic (Levy and Marshall 2004) , the key manage ment aim seems to be to delay and confine such a development as much as possible.

Second, a robust quantitative risk assess ment will require rates of HGT and/or gene mutations in the relevant compartments ( Figure 1 , processes 3-5) to be described for different combinations of donating eARB strains and receiving pARB strains. The lack of quantitative estimates for mutation/HGT of ARG is a major data gap.

Third, traditional microbial risk assess ment dose-response approaches (Figure 1, processes 6 and 8) could be used to address the likeli hood of infection [Codex Alimentarius Commission 2011; U.S. EPA and U.S. Department of Agriculture/Food Safety and Inspection Service (USDA/FSIS) 2012], but the novel aspect required here-in addition to HGT and ARB selection-would be to address quantitative dose-response relation ships for eARB (in the presence of a sensitive pathogen in or on a human) (Figure 1 , pro cesses 3 and 6). Importantly, the key difference from traditional HHRA undertaken in some jurisdictions is that it is essential to include environmental processes to fully assess human risks associated with anti biotic resistance.

Therefore, the type of information that should be documented for a human healthoriented risk assessment of environmental ARB includes the following [adapted from Codex Alimentarius Commission (2011)]:

• Clinical and environmental surveillance programs for anti biotics, ARB, and their determinants, with a focus on regional data volume 121 | number 9 | September 2013 • Environmental Health Perspectives reporting the types and use of anti biotics in human medicine, crops, and commercial and companion animals, as well as globally where crops and food animals are produced • Epidemiological investigations of outbreaks and sporadic cases associated with ARB, including clinical studies on the occurrence, frequency, and severity of ARB infections • Identification of the selection pressures (time and dose of selecting/coselecting agents) required to select for resistance in differ ent environments, and subsequent HGT to humanrelevant bacteria, both based on reports describing the frequency of HGT and uptake of ARG into environmental bac teria, including environmental pathogens, in previously identified hot spots • Human, laboratory, and/or field animal/crop trials addressing the link between anti biotic use and resistance (particularly regional data) • Investigations of the characteristics of ARB and their determinants (ex situ and in situ) • Studies on the link between resistance, viru lence, and/or ecological fitness (e.g., surviv ability or adaptability) of ARB • Studies on the environmental fate of anti biotic residues in water and soil and their bioavailability associated with the selection of ARB in any given environmental com partment, animal, or human host result ing in pARB • Existing risk assessments of ARB and related pathogens.

In summary, many sources of data are required to undertake a human health risk assessment for environ mental ARB, and much of the data may be severely limited (particularly for a quantitative assessment). Thus, the final risk assessment report should emphasize the importance of the evidence trail and weight of evidence for each finding. Furthermore, when models are constructed, previously unused data sets should be consid ered for model verifications where possible.

Human health risk assessment of anti biotics in the environment builds on traditional chemical risk assessments (National Research Council 1983), starting, for example, with an accept able daily intake (ADI) based on resistance data (VICH Steering Committee 2012). A corresponding metric for environ mental anti biotic concentration could be developed based on the concept of the minimum selective concentration (MSC) (Gullberg et al. 2011) , defined as the minimum concentration of an anti biotic agent that selects for resistance. Unlike the traditional chemical risk assess ment approach, with the MSC assay it would be necessary to address the human health effects arising from ARGs and the resistance determinants that give rise to ARB, including resistance associated with mutations (Figure 1 , processes 1 and 2). In the absence of specific data, an MSC assay could inform a risk asses sor of the selective concentration of a pharma ceutical or complex mixture of compounds in a matrix of choice, allowing description of thresholds for significant ARB development.

Pathogen risks may be evaluated through microbial risk assessment (MRA), a struc tured, systematic, sciencebased approach that builds on the chemical risk assessment paradigm; the MRA involves a) problem for mulation (describing the hazards, risk setting, and pathways), b) exposure assessment of the hazard (ARB, ARG), c) dose-response assess ment that quantifies the relationship between hazard dose and pARB infection in humans (Figure 1, processes 6 and 7) , and d) com bination of these procedures to characterize risk for the various pathways of exposure to pathogens identified to be assessed. An MRA is used qualitatively or quantitatively to evalu ate the level of exposure and subsequent risk to human health from microbiological haz ards. In the context of anti bioticresistant micro organisms, environmental MRA is in its infancy but is needed to address resistant bac teria and/or their determinants. The MRA was originally developed for fecal pathogen hazards in food and water [ILSI (International Life Sciences Institute) 1996], with more recent modifications to include biofilmassociated environmental pathogens such as Legionella pneumophila (Schoen and Ashbolt 2011) . Some human pathogens can grow in the envi ronment (and may become pARB; Figure 1 , processes 1 and 2), and many will infect only compromised individuals (generally termed opportunistic pathogens).

Over the past 20 years, the MRA has largely evolved by input from the inter national food safety community, and it is now a wellrecognized and accepted approach for food safety risk analysis. In 1999, the Codex Alimentarius adopted the Principles and Guidelines for the Conduct of Microbiological Risk Assessment (CAC/GL30) (Codex Alimentarius Commission 2009). The most recent Codex Alimentarius guidelines for risk analysis of foodborne antimicrobial resistance include eight principles (Codex Alimentarius Commission 2011), and in the United States, MRA guidelines for food and water (U.S. EPA and USDA/FSIS 2012) continue to use the fourstep framework originally described for chemical risk assessment. Several ARB risk assessments have been published and reviewed in recent years (Geenen et al. 2010; McEwen 2012; Snary et al. 2004) . However, nearly all of these studies focus on foodborne transmis sion; human health risk assessments dealing with ARB transmission via various environ mental routes or direct contact with ARG are sparse.

For example, Geenen et al. (2010) studied extendedspectrum betalactamase (ESBL) producing bacteria and identified the following risk factors: previous admission to healthcare facilities, use of anti microbial drugs, travel to highendemic countries, and having ESBL positive family members. The authors con cluded that an environ mental risk assessment would be helpful in addressing the problem of ESBLproducing bacteria but that none had been performed.

Hazard identification and hazard charac terization. Unfortunately, we are unaware of data that quantitatively link ARG uptake and human health effects (Figure 1 , processes 3 and 6). What data do exist and are rapidly improving in quality, however, are on the presence of ARGs within various environ mental compartments (Allen et al. 2009; Cummings et al. 2011; Ham et al. 2012) , specifically on clinically rele vant resistance genes within soils (Forsberg et al. 2012 ) (Figure 1 , process 1). Precursors that lead to the develop ment of ARB hazards include ARG and mecha nisms to mobilize these genes, anti biotics, and coselecting agents (Qiu et al. 2012; Taylor et al. 2011 ) along with gene mutations (Gillings and Stokes 2012) . Depending on the presence of recipient bac teria, these processes generate either eARB or pARB (Figure 1, processes 1 and 2) .

In regard to the numerous parameters rele vant to individual environmental compart ments, we are not aware of the availability of comprehensive data on a) anti biotic resistance development by anti biotics and other coselect ing agents; b) the flow of ARG (resistome) and acquisition elements (e.g, integrons) in native environmental compartment bacteria; or c) the likely range in rates of horizontal and vertical gene transfer within environ mental compartments. Nonetheless, factors that are considered important include the range of potential pathways involving the release of anti biotics, ARG, and ARB into and amplify ing in environmental compartments such as the rhizosphere, bulk soil, compost, biofilms, wastewater lagoons, rivers, sedi ments, aqua culture, plants, birds, and wildlife.

With respect to anti biotics, in general, the following information is required to aid haz ard characterization: a) a list of the local anti biotic classes of concern, b) what is known of their environmental fate, and c) where they may accumulate, in particular, environmental compartments (e.g., the rhizosphere, general soil, compost, biofilms, wastewater lagoons, rivers, sediments, aquaculture, plants, birds, wildlife, farm animals, or companion ani mals). Selection for ARB (Figure 1 , process 2) will depend on the type and in situ bio availability of selecting/coselecting agents, the abundance of bacterial host, and the abun dance of AR determinants.

Selection for ARB is further modulated by the nutritional status of members of the rele vant bacterial community because high meta bolic activity and high cell density promote bacterial community succession and HGT (Brandt et al. 2009; Sørensen et al. 2005) . In contrast, HGT is relatively independent of anti biotics-although anti biotics and ARB may be cotransported (Chen et al. 2013 )and increases in HGT rates are thought to occur in stressed bacteria. For example, integrase expression can be upregulated (increased) by the bacterial SOS response (process for DNA repair) in the presence of certain anti biotics (Guerin et al. 2009 ).

Although quantitative data that describe the development of pARB in the environment are lacking, ample evidence exists for the co uptake by an antibioticsensitive pathogen in the presence of an anti biotic, ARG (such as on a plasmid with metal resistance), and/or carbon utilization genes (Knapp et al. 2011; Laverde Gomez et al. 2011) , or as demon strated in vitro for a disinfectant/nanomaterial (Qiu et al. 2012; Soumet et al. 2012) .

The spatial distribution of organisms (opportunity for close proximity) may also affect gene transfer, which results from inher ent patchi ness, soil structure, presence of substrates, and so forth. In considering gene transfer rates, there may be hot spots of activ ity; for example, there is evidence for HGT of clinically rele vant resistance genes between bacteria in manureimpacted soils (Forsberg et al. 2012 ) and in association with the rhi zosphere because of its organicrich condi tions (Pontiroli et al. 2009 ). In addition, selection pressures for subsequent prolifera tion of eARB may be higher in these hot spot environments (Brandt et al. 2009; Li et al. 2013) . Therefore, it is important to reco gnize likely zones of high activity during the prob lem formulation and hazard characterization stages of a risk assessment, and when using sampling to identify in situ exchange rates. As an example marker of anthropogenic impact with potential to predict the impact on the mobile resistome, class 1 integrons could be used because of their ability to integrate gene cassettes that confer a wide range of anti biotic and biocide resistance (Gaze et al. 2011) . In semipristine soils, prevalence may be two or three orders of magnitude lower than in impacted soils and sedi ments (0.001 vs. 1%, respectively) (Gaze et al. 2011; Zhu et al. 2013) .

In addition to a huge diversity of eARB hazards, there are several pathogens that could be evaluated in microbial risk assess ments: a) foodborne and waterborne fecal pathogens represented by Campylobacter jejuni, Salmonella enterica, or various patho genic E. coli; and b) environ mental pathogens, such as respiratory, skin, or wound pathogens represented by Legionella pneumophila, Staphylococcus aureus, and Pseudomonas aeruginosa. Each of these fecal and environmental pathogens is well known to be able to acquire ARG; thus, given further data on environmen tal HGT rates, they could be used as refer ence pathogens in microbial risk assessments. However, what is much more problematic for risk assessment-and a current limiting factor-is the rate at which the indigenous bacteria transfer resistance to these pathogens within each environmental compartment and within the human/animal host (Figure 1 , pro cesses 3-5). Methods to model and experi mentally derive relevant information on these environmental issues are discussed below in "Environmental Exposure Assessment." Data on HGT within the human gastro intestinal tract have been summarized by Hunter et al. (2008) .

Dose-response relationships. To properly charac terize human risks, it is typical to select hazards for which there are dose-response health data described either deterministically or stochastically, as available for the refer ence enteric pathogens (e.g., Campylobacter jejuni, Salmonella enterica, E. coli) (Schoen and Ashbolt 2010) , but these dose-response health data have yet to be quantified for the skin/wound reference pathogens (Mena and Gerba 2009; Rose and Haas 1999) . However, as noted above for processes 1-5, (Figure 1 ), an important difference for ARB is the need to account for the phenomena associated with selective environmental pressures for the development of ARB, and ultimately that form the human infective dose of either eARB or pARB. The exact mechanisms and doseresponse relationships have yet to be eluci dated, and may be different depending on the bacterial species and resistance mechanisms involved. Nevertheless, it seems reasonable for the non compromised human exposed to a pARB to fit the published dose-response/ infection relationship (e.g., derived from "feeding" trials with healthy adults or from information collected during outbreak inves tigations) for strains of the same pathogen without antibiotic resistance. What appears more limiting are dose-response models that describe the probability of illness based on the conditional probability of infection and including people who are already compro mised, such as those under going anti biotic therapy. Although there is definitive data on pARB being more pathogenic or causing more severe illness than their antimicrobial susceptible equivalents (Barza 2002; Helms et al. 2004 Helms et al. , 2005 Travers and Barza 2002) , that may not always be the case (Evans et al. 2009; Wassenaar et al. 2007) . Clear examples of excess mortality include associ ated blood stream infections for methicillin resistant Staphylococcus aureus (MRSA) and from third generation cephalosporinresistant E. coli (G3CREC). In 2007 in participating European countries, 27,711 cases of MRSA were associated with 5,503 excess deaths and 255,683 excess hospital days, and 15,183 epi sodes of G3CREC blood stream infections were responsible for 2,712 excess deaths and 120,065 extra hospital days (de Kraker et al. 2011) . The authors predicted that the combined burden of resistance of MRSA and G3CREC will likely lead to a pre dicted incidence of 3.3 associated deaths per 100,000 inhabitants in 2015. Yet for many regions of the world, such predictions are less well understood.

The final step of MRA is risk charac teriza tion, which integrates the outputs from the hazard identification, the hazard charac terization, dose response, and the exposure assessment with the intent to generate an overall estimate of the risk. This estimate may be expressed in various measures of risk, for example, in terms of individual or popula tion risk, or an estimate of annual risk based on exposure to specific hazard(s). Depending on the purpose of the risk assessment, the risk characterization can also include the key scientific assumptions used in the risk assessment, sources of variability and uncer tainty, and a scientific evalua tion of risk management options.

Based on our conceptualization of the pro cesses important to undertake HHRA of ARB (Figure 1 ), most elements related to ARB development in environmental media (pro cesses 1, 2, and 4) have been addressed above in "Hazard identification and hazard charac terization." Here we focus on describing important environmental compartments for and human exposure to ARB (Figure 1 , pro cesses 3 and 6). Concentrations of environ mental factors (such as anti biotics) and ARB, along with their fate and transport to points of human uptake, are critical to exposure assessment. For a particular human health risk assessment of ARB, it would be impor tant to select/expand on individual pathway scenarios (identifying critical environmental compartments to human contact) relevant to the anti biotic/resistance determinants identi fied in the problem formulation and hazard characterization stages.

Compartments of potential concern include soil environments receiving animal manure or biosolids, compost, and lagoons, rivers, and their sediments receiving waste waters (Chen et al. 2013 ). More traditional routes of human exposures to contaminants that could include eARB and pARB are drinking water, recreational and irrigation waters impacted by sewage and/or anti biotic volume 121 | number 9 | September 2013 • Environmental Health Perspectives production wastewaters, food, and air affected by farm buildings and exposure to farm ani mal manures, as discussed by Pruden et al. (2013) . What is emerging as an important research gap is the in situ development of ARB within biofilms (Boehm et al. 2009) and their associated freeliving protozoa that may protect and transport ARB (Abraham 2010) to and within drinking water systems (Schwartz et al. 2003; Silva et al. 2008 ). This latter route could be particularly problem atic for hospital drinking water systems where an already vulnerable population is exposed. In addition, with the increasing use and exposure to domestically collected rainwa ter, atmospheric fallout of ARB may "seed" household systems (Kaushik et al. 2012) .

After identifying anti biotic concentra tions and pathogen densities in the environ ment, as well as possible levels and rates of ARB generation in each environmental compartment, a range of fate and transport models are available to estimate the amounts of anti biotics, pathogens, ARB, and ARG at points of human contact (Figure 1 , pro cesses 3 and 6). Such models are largely based on hydro dynamics, with pathogenspecific parameters to account for likely inactivation/ predation in soil and aquatic environments, such as sunlight inactiva tion (Bradford et al. 2013; Cho et al. 2012; Ferguson et al. 2010) .

A key aspect of the fate and transport models is the ability to account for the inherent vari ability of any system component. In addition, our uncertainties in assessing model parameter values should be factored into fate and trans port models such as by using Bayesian syn thesis methods (Albert et al. 2008; Williams et al. 2011) . To better account for param eter uncertainties, more recent models are including Bayesian learning algorithms that help to integrate information using meteo rologic, hydrologic, and microbial explana tory variables (Dotto et al. 2012; Motamarri and Boccelli 2012) . Overall, these models also help to identify management opportunities to mitigate exposures to ARB and anti biotics and are an important aspect in describing the path ways of hazards to points of human exposure in any risk assessment.

Considering the complexity of exposure path ways associated with environmental ARB risks and the large uncertainty in the input data for some nodes along the various exposure path ways, outputs would inevitably be difficult for decision makers to interpret and could in fact be counter productive. Thus, there is merit in considering decision analysis approaches to prioritize risks, guide resource allocation and data collection activities, and facilitate decision making. Although there is a range of ranking options, uses of weightings, and selection criteria (Cooper et al. 2008; Pires and Hald 2010) , as well as failure mode and effects analysis (Pillay and Wang 2003) , in the overall area of microbial risk assessment, there is a consolidation to MCDA approaches that may include Bayesian algorithms (Lienert et al. 2011; Ludwig et al. 2013; Ruzante et al. 2010) .

Approaches such as MCDA are designed to provide a structured framework for mak ing choices where multiple factors need to be considered in the decisionmaking pro cess. MCDA is a wellestablished tool that can be used for evaluating and document ing the importance assigned to different fac tors in ranking risks (Lienert et al. 2011) , albeit heavily dependent on expert opinion. In the context of MRA, MCDA has been used to rank foodborne microbial risks based on multiple factors, including public health, market impacts, consumer perception and acceptance, and social sensitivity (Ruzante et al. 2010) , as well as to prioritize and select inter ventions to reduce pathogen exposures (Fazil et al. 2008) . Examples of MCDA applications in structuring decisions for man aging eco toxico logi cal risks have also been reported (Linkov et al. 2006; Semenzin et al. 2008 ) and provide useful MCDA approaches. MCDA could be used, for example, to evalu ate and rank the relative risks between habi tats highly polluted with anti biotics, ARG, and ARG determinants, as described above for possible hot spots for HGT and develop ment of ARB. MCDA could be applied to evaluate the relative contribution of coselect ing agents (e.g., detergents, biocides, met als, nano materials) from various sources to the overall risk of ARB in the environment. Moreover, for a range of anti biotics consid ered to be of environmental concern, MCDA approaches could be used for risk ranking according to criteria based on relevant con tributing factors (e.g., mobility of resistance determinants in genetic elements, antibiotic resistance transfer rates in different environ mental compartments, accumulation levels of anti biotics in environmental compartments, environmental fate and transport to expo sure points). In the MCDA process, it is also important to identify low probability but high impact "onetimeevent" types of risk.

Because MCDA techniques rely on expert opinion (which is often regarded as a limi tation of such approaches), wellstructured and recognized elicitation practices should be used in order to avoid introduction of biases and errors by subjective scoring. In contrast, one of the main advantages of MCDA tech niques is that they capture a consensus opin ion among an expert group about the most relevant criteria and their relative weight on the decision.

There are several research gaps that need to be addressed. In particular, specific atten tion should be paid to contaminated habitats (hot spots) in which anti biotics, coselecting agents, bacteria carrying resistance determi nants on mobile genetic elements, and favor able conditions for bacterial growth and activity-all conditions expected to favor HGT-prevail at the same time. However, because these data are currently very limited, workshop participants evaluated alternative ways and possible experimental methods to address these data gaps for HHRA as they relate to the processes identified in Figure 1 .

Assays to determine MSC (processes 1, 2,  and 4) . Assays could be developed to mea sure MSC (Gullberg et al. 2011 ) for a range of anti biotics and environmental conditions. For example, assays could be developed and validated in sandy and clay soils, different sediments, and water types, with isogenic pairs of the model organism inoculated into the matrix of choice and subjected to a titra tion of the selective agent to sufficiently high dilution. Selection at sub inhibitory concen trations and assay development for environ mental matrices are key areas of research that need to be addressed. However, overall care is needed when interpreting ex situ studies and extrapolating to in situ environmental condi tions, as well as in dealing with illdefined hazard mixtures in the environment. (processes 1, 2, and 4) . Hot spots, locations where a highlevel of HGT and anti biotic resistance develop, may, for instance, include aquatic environments affected by pharma ceutical industry effluents, aqua culture, or sewage discharges, as well as terrestrial environments affected by the deposition of biosolids or animal manures. The degree of persistence of anti biotic resistance (i.e., the rate by which resistance disappears without having an environ mental selection pressure for resistance) must also be considered for risk assessment and will depend on the fit ness cost of resistance. However, the fitness costs within complex and variable environ ments are difficult to assess. Furthermore, standard methods have not been developed for evaluating environ mental selection pres sures in complex microbial communities, but several experimental approaches are possible and have been described elsewhere (Berg et al. 2010; Brandt et al. 2009 ).

The approaches identified by Berg et al. (2010) and Brandt et al. (2009) could be labo ra tory based (to assess the potency of known compounds/mixtures) or applied in the field to assess whether the environment in question (with, for example, its unknown mixture of chemicals) is a hot spot. Defining "critical exposure levels" is therefore an important HHRA output to aid manage ment activities, which will likely vary between and within environmental compartments, depending on the location.

Screening for novel resistance determinants (to reduce process 2). Screening procedures could be introduced early in the development cycle of novel anti biotics to ensure that exist ing resistance determinants are not prevalent in environmental compartments. Marked recipient strains could be inoculated into environmental matrices [e.g., soil, biosolids, or fecal slurry (with sterilized matrix equiva lents as negative controls)], incubated, and then seeded onto media containing the study compound along with a selective anti biotic to recover marked recipient strains demon strating resistance. Plasmids, or the entire genome of the recipient, could then be cloned into small insert expression vectors, transformed into E. coli or other hosts, and seeded back onto media containing the study compound. In this way, novel resistance determinants would be identified.

Alternatively, functional meta genomics could be used to identify novel resistance determinants in meta genomic DNA (Allen et al. 2009 ). In brief, DNA would be extracted from an environmental sample, cloned into an expression vector, and trans formed into a bacterial host such as E. coli. Transformants could then be screened on the study compound and resistance genes identi fied using transposon muta genesis followed by sequencing and bio informatic analyses. This would allow detection of novel resistance determinants that may not be plasmid borne but may transfer to other pathogens.

Dose-response data needs (processes 3, 5, and 6). We were unaware of dose-response data representing a combined ARG and a recipient, previously susceptible pathogen dose, and human or animal disease (Figure 1 , processes 3 and 5). In contrast, various exam ples illustrate increased morbidity and mor tality when humans are exposed to pARB, as described above in "Dose-response rela tionships." Hence, existing published doseresponse models for non resistant pathogens may not be appropriate to use beyond the end point of infection, and further dose-response models that address people of various lifestages need to be described and summarized to facilitate pARB risk assessments. There is also a need to develop dose-response information for sec ondary illness end points (sequelae) resulting from pARB infections.

Regarding anti biotic concentration and time of exposure giving rise to selection of pARB within a human (for couptake of eARB and a sensitive pathogen), safety could be based on the effective concentration for the specific anti biotic under consideration. In other words, screening values to determine whether further action is warranted could be derived from the acute or mean daily anti biotic intake, with uncertainty factors applied as appropriate, until future research is under taken on pathogen anti bioticresponse changes in the presence of specific anti biotic treatment. Alternatively, epidemiological data from exist ing clones of anti bioticresistant strains (e.g., NDM1) could provide useful data for doseresponse and exposure assessments.

Options for ranking risks (overall HHRA). In the absence of fully quantitative data to undertake a HHRA, riskranking approaches based on exposure assessment modeling could be adopted and developed to inform allocation of resources for data generation as part of an HHRA of ARB. Evers et al. (2008) presented one such approach in the context of food safety for estimating the relative contribution of Campylobacter spp. sources and transmis sion routes on exposure per personday in the Netherlands. Their study included 31 transmis sion routes related to direct contact with animals and ingestion of food and water, and resulted in a ranking of the most significant sources of exposure. Although their study focused on foodborne transmission routes and did not deal with anti bioticresistant Campylobacter strains, a similar approach could be applied to estimate human exposure to ARB hazards using the environmental exposure pathways described by Evers et al. (2008) . This would require data on the prevalence of ARG and ARB, as well as lev els of anti biotics present in all exposure routes to be considered in the risk assessment. Although such an approach is probably not currently fea sible, improved environmental data for a select number of pathogen-gene combinations could be developed in the future.

An alternative approach to capturing knowledge of experts and other stakeholders could be to develop a Bayesian network based on expert knowledge and add to that as data become available, as described for campylo bacters in foods by Albert et al. (2008) .

Because we are addressing an inter national problem and because the precautionary approach is used in many jurisdictions, there are many risk management approaches that can be implemented now, before anti biotic resistance issues worsen, as noted in the related risk management paper resulting from the workshop (Pruden et al. 2013) . Furthermore, many current risk management schemes start the process from a management perspec tive and delve into quantitative assessments as needed in order to improve risk manage ment actions, such as in the WHO water safety plans (WHO 2009) . We propose that environmental aspects of anti bioticresistance development be included in the processes of any HHRA addressing ARB. In general terms, an MRA appears suitable to address environ mental human health risks posed by the envi ronmental release of anti biotics, ARB, and ARG; however, at present, there are still too many data gaps to realize that goal. Further development of this type of approach requires data mining from previous epidemiological studies to aid in model development, param eterization, and validation, as well as in the collection of new information, particularly that related to conditions and rates of ARB development in various hot spot environ ments, and for various human health doseresponse unknowns identified in this review. In the nearterm, options likely to provide a firstpass assessment of risks seem likely to be based on MCDA approaches, which could be facilitated by Bayesian network models. Once these MRA models gain more acceptance, they may facilitate scenario testing to deter mine which control points may be most effec tive in reducing risks and which riskdriving attributes should be specifically considered and minimized during the development of novel anti biotics.

Megacities as sources for pathogenic bacteria in rivers and their fate downstream

Quantitative risk assessment from farm to fork and beyond: a global Bayesian approach concerning food-borne diseases

Functional metagenomics reveals diverse betalactamases in a remote Alaskan soil

Antibiotic resistance and its cost: is it possible to reverse resistance?

Persistence of antibiotic resistance in bacterial populations

Positive and negative selection towards tetracycline resistance genes in manure treatment lagoons

Potential mechanisms of increased disease in humans from antimicrobial resistance in food animals

Cu exposure under field conditions coselects for antibiotic resistance as determined by a novel cultivationindependent bacterial community tolerance assay

Second messenger signalling governs Escherichia coli biofilm induction upon ribosomal stress

Quantification of genes encoding resistance to aminoglycosides, beta-lactams and tetra cyclines in wastewater environments by real-time PCR

Transport and fate of microbial pathogens in agricultural settings

Environmental Health Perspectives community tolerance to sulfadiazine in soil hotspots amended with artificial root exudates

Multiresistance, beta-lactamase-encoding genes and bacterial diversity in hospital wastewater in Rio de Janeiro, Brazil

Fate and transport of antibiotic residues and antibiotic resistance genes following land application of manure waste

Differentiating anthropogenic impacts on ARGs in the Pearl River Estuary by using suitable gene indicators

Class 1 integrons, selected virulence genes, and antibiotic resistance in Escherichia coli isolates from the Minjiang River

The modified SWAT model for predicting fecal coliforms in the Wachusett Reservoir Watershed, USA

Principles and Guidelines for the Conduct of Microbiological Risk Assessment. CAC/ GL-30

Guidelines for Risk Analysis of Foodborne Antimicrobial Resistance

The role of drinking water in the transmission of antimicrobial-resistant E. coli

Preliminary risk assessment database and risk ranking of pharmaceuticals in the environment

Persistence of transferable extended-spectrum-β-lactamase resistance in the absence of antibiotic pressure

The soil resistome: the anthropogenic, the native, and the unknown

Broad dissemination of plasmidmediated quinolone resistance genes in sediments of two urban coastal wetlands

Antibiotic resistance is ancient

ESBL-positive enterobacteria isolates in drinking water

Mortality and hospital stay associated with resistant Staphylococcus aureus and Escherichia coli bacteremia: estimating the burden of antibiotic resistance in Europe

CTX-M-15-producing Escherichia coli clone B2-O25b-ST131 and Klebsiella spp. isolates in municipal wastewater treatment plant effluents

Comparison of different uncertainty techniques in urban stormwater quantity and quality modelling

Short-term and medium-term clinical outcomes of quinolone-resistant Campylobacter infection

Campylobacter source attribution by exposure management

Choices, choices: the application of multi-criteria decision analysis to a food safety decision-making problem

Modeling of Variations in Watershed Pathogen Concentrations for Risk Management and Load Estimations

Contamination of surface, ground, and drinking water from pharmaceutical production

The scourge of antibiotic resistance: the important role of the environment

United Nations/World Health Organization/ World Organisation for Animal Health

The shared antibiotic resistome of soil bacteria and human pathogens

Correlation of tetra cycline and sulfonamide antibiotics with corresponding resistance genes and resistant bacteria in a conventional municipal waste water treatment plant

Impacts of anthropogenic activity on the ecology of class 1 integrons and integron-associated genes in the environment

Risk profile on Antimicrobial Resistance Transmissible from Food Animals to Humans. RIVM rapport 330334001. Bilhoven:National Institute for Public Health and the Environment (RIVM)

Are humans increasing bacterial evolvability?

A framework for global surveillance of antibiotic resistance

The SOS response controls integron recombination

Selection of resistant bacteria at very low antibiotic concentrations

Quantitative Microbial Risk Assessment

Distribution of antibiotic resistance in urban watershed in Japan

Quinolone resistance is associated with increased risk of invasive illness or death during infection with Salmonella serotype typhimurium

Adverse health events associated with antimicrobial drug resistance in Campylobacter species: a registry-based cohort study

Metaanalysis of experimental data concerning antimicrobial resistance gene transfer rates during conjugation

A conceptual framework to assess the risks of human disease following exposure to pathogens

bla NDM-1 -positive Klebsiella pneumoniae from environment

Influence of air quality on the composition of microbial pathogens in fresh rainwater

Antimicrobial Resistance in the Environment

Antibiotic resistance gene abundances correlate with metal and geochemical conditions in archived Scottish soils

Pyrosequencing of antibioticcontaminated river sediments reveals high levels of resistance and gene transfer elements

Antibiotics in the aquatic environment-a review-part I

Antibiotics in the aquatic environment-a review-part II

Effluent from drug manufactures contains extremely high levels of pharmaceuticals

A multiresistance megaplasmid pLG1 bearing a hyl Efm genomic island in hospital Enterococcus faecium isolates

Antibacterial resistance worldwide: causes, challenges and responses

Antibioticresistance profile in environmental bacteria isolated from penicillin production wastewater treatment plant and the receiving river

Antibiotic resistance characteristics of environmental bacteria from an oxytetracycline production wastewater treatment plant and the receiving river

Occurrence of chloramphenicol-resistance genes as environmental pollutants from swine feedlots

Multiple-criteria decision analysis reveals high stakeholder preference to remove pharmaceuticals from hospital wastewater

From comparative risk assessment to multi-criteria decision analysis and adaptive management: recent developments and applications

Identifying associations in Escherichia coli antimicrobial resistance patterns using additive Bayesian networks

Quantitative human health risk assessments of antimicrobial use in animals and selection of resistance: a review of publicly available reports

Risk assessment of Pseudomonas aeruginosa in water

Development of a neural-based forecasting tool to classify recreational water quality using fecal indicator organisms

Modified failure mode and effects analysis using approximate reasoning

Assessing the differences in public health impact of Salmonella subtypes using a Bayesian microbial subtyping approach for source attribution

Visual evidence of horizontal gene transfer between plants and bacteria in the phytosphere of transplastomic tobacco

Management options for reducing the release of antibiotics and antibiotic resistance genes to the environment

Nanoalumina promotes the horizontal transfer of multiresistance genes mediated by plasmids across genera

A risk assessment framework for the evaluation of skin infections and the potential impact of antibacterial soap washing

A multifactorial risk prioritization framework for foodborne pathogens

Assessing pathogen risk to swimmers at non-sewage impacted recreational beaches

An in-premise model for Legionella exposure during showering events

Detection of antibiotic-resistant bacteria and their resistance genes in wastewater, surface water, and drinking water biofilms

Integration of bioavailability, ecology and ecotoxicology by three lines of evidence into ecological risk indexes for contaminated soil assessment

Prevalence of antibiotic resistance genes in the bacterial flora of integrated fish farming environments of Pakistan and Tanzania

Metagenomic insights into chlorination effects on microbial antibiotic resistance in drinking water

Characterisation of potential virulence markers in Pseudomonas aeruginosa isolated from drinking water

Antimicrobial resistance: a microbial risk assessment perspective

Studying plasmid horizontal transfer in situ: a critical review

Resistance to phenicol compounds following adaptation to quaternary ammonium compounds in Escherichia coli

Wildlife trade and the emergence of infectious diseases

Aquatic systems: Maintaining, mixing and mobilising antimicrobial resistance?

Morbidity of infections caused by antimicrobial-resistant bacteria

Environmental Protection Agency). 2012. Human Health Risk Assessment

Microbial Risk Assessment Guideline: Pathogenic Microorganisms with Focus on Food and Water. EPA/100/J-12/001

Diversity and antibiotic resistance patterns of Sphingomonadaceae isolates from drinking water

Studies to Evaluate the Safety of Residues of Veterinary Drugs in Human Food: General Approach to Establish a Microbiological ADI. VICH GL36(R)

Re-analysis of the risks attributed to ciprofloxacin-resistant Campylobacter jejuni infections

Water Safety Plan Manual: Step-by-Step Risk Management for Drinking-Water Suppliers. Geneva:World Health Organization

2012a. Report of the 3rd Meeting of the WHO Advisory Group on Integrated Surveillance of Antimicrobial Resistance

The Evolving Threat of Antimicrobial Resistance: Options for Action. Geneva:World Health Organization

Antimicrobial Resistance. Fact sheet No. 194

Framework for microbial food-safety risk assessments amenable to Bayesian modeling

NDM-1 and the role of travel in its dissemination

Diverse and abundant antibiotic resistance genes in Chinese swine farms