Identifying and Reducing Environmental Health Risks of Chemicals in Our Society: Workshop Summary (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

5 Approaches to Prioritizing Chemicals for Risk Assessment and Risk Management Roundtable member and session moderator Andrew Maguire opened the fourth session and suggested that the overall theme of the session was innovation. “We see a bewildering array of levels and types of what we know or think we know and also what we don’t know,” he said. “We see that there are very long cycles for decisions. Is the data relevant enough or not? Is it sufficiently conclusive or not for policy making? . . . For this panel, the question is, What can we do now? What can we do soon enough to protect the health of many millions of people? How do we make the decisions that we need to make? How do we move ahead?” In particular, the topic of the session was examples of improved approaches to priority setting in the risk assessment and risk management of chemicals. A good name for the panel, Maguire suggested, would be the Who’s Doing What Panel, as each panel member was to describe what his or her institution is doing to improve the prioritization of chemicals for testing and management. The four institutions represented on the panel were the U.S. Environmental Protection Agency (EPA), the California Environmental Protection Agency (California EPA), Health Canada, and the American Chemistry Council (ACC). DEVELOPING MODELS TO PRIORITIZE CHEMICALS FOR TARGET TESTING The session’s first presenter was Richard Judson of the National Center for Computational Toxicology at the EPA. He spoke about the use of toxicological models to prioritize chemicals for testing. 73 

74 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS Judson began by noting that although he had been asked to speak about ToxCastTM,1 the EPA’s program for forecasting toxicity, his talk would actually cover a much larger collaborative effort among the EPA and a number of other groups. “We have collaborators at the National Toxicology Program and elsewhere at the NIH [National Institutes of Health], the FDA [Food and Drug Administration], and a number of academic and government and industry collaborators, both here and in Europe,” he said. “It is a very big collaboration where we are all working on different parts of this problem of solving the too many chemicals problem. It is not just us.” The EPA’s computational toxicology center was founded nine years ago to deal with the “too many chemicals” problem. “We know that there are lots of chemicals out there,” he said. “Most of them have never been tested in the standard animal tests. A critical issue is: How can we take thousands of chemicals and run a consistent set of tests on them?” A second issue is how to use in vitro methods combined with the results from tests on laboratory animals to infer the human risks of those chemicals. “We and our collaborators have come up with an overall strategy,” he said. The basic idea is to develop a better understanding of the modes of action of these chemicals in the human body and thus to be able to use data from tests on laboratory animals and elsewhere to build models that can be used to predict computationally the effects of various chemicals on humans. There are four basic steps in this strategy: to develop high- throughput in vitro assays for testing chemicals on the biological pathways linked to toxicity, to use that information to develop predictive hazard models, to develop high-throughput exposure predictions, and to create the data and models to assess risk from the chemicals. These models can be used to prioritize chemicals for targeted testing, to distinguish possible adverse outcome pathways for chemicals, and to provide semiquantitative high-throughput risk assessments. Judson then went into detail on each of these steps in the computational toxicology approach. Before beginning, it is necessary to identify the various biochemical pathways through which chemicals exert their influence in the human body. Generally speaking, the chemicals work by perturbing the usual function of a pathway in some way that leads to a toxic effect. Thanks to a great deal of work by scientists in a variety of fields, many of these pathways have already been identified. 1 Further information on ToxCastTM is available at http://www.epa.gov/ncct/ toxcast (accessed March 31, 2014). 

APPROACHES TO PRIORITIZING CHEMICALS 75 With this information in hand, the next step in the program is to develop high-throughput in vitro assays to test for the effects of various chemicals on these biochemical pathways. The effects of the chemicals can be tested on proteins or in whole cells or even in something as complicated as zebrafish. “You can actually treat whole zebrafish in a microtiter plate well,” Judson said. He noted that researchers can detect a variety of phenotypes using this approach. In particular, zebrafish provide a good model for detecting chemicals causing a range of developmental defects. Currently, several laboratories have tested thousands of chemicals in these sorts of assays. The EPA did not have to pay for most of the technology development for the assays, Judson said, because most of the development had already been done by the pharmaceutical industry. “They spend a billion dollars developing every drug,” he said. “They have made investments in basic technology, which we were then able to buy off the shelf.” There are many different types of assays that can be run, Judson said. “You can use models—you just start with structure and run it on the computer. There are assays that are run in cells, assays run in zebrafish, or in C. elegans.” At one extreme, he explained, we can run cell-based assays that tell us what is going on with every one of the 20,000 genes in the whole genome. At the moment, he said, these whole-genome assays are too expensive—thousands of dollars per chemical—to do for every chemical. However, “The Broad Institute, 2 weeks from now, is releasing its first big dataset with a version of this chip that is getting down to hundreds of dollars per chemical. That puts doing whole-genome analysis for every chemical that we are exposed to within the realm of possibility.” Running hundreds of assays on thousands of chemicals results in huge datasets. The largest dataset that has been generated in the cross- agency Tox212 collaboration has about 8,200 chemicals in it, including both environmental chemicals and chemicals found in consumer products (Zang et al., 2013). There are a number of flame retardants and also drugs, food additives, chemicals in consumer products that are put on the skin, and so forth. With those data in hand, the next step is to create models that can accurately predict what happens when people are exposed to the chemicals. 2 Further information on Tox21 is available at http://epa.gov/ncct/Tox21 (accessed March 31, 2014). 

76 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS The most basic approach is to take in vitro data and in vivo data and apply statistical analyses to look for relationships between the two sets of data; ideally, the relationships would make it possible to use in vitro data to predict the effects on humans of chemicals for which there are no in vivo data. “In the most basic approach, you treat all of this in vitro data as just a bunch of numbers,” Judson explained. “The chemical either turns on that pathway or not. Likewise, you treat all of the in vivo data as just another bunch of numbers. This chemical either causes cancer or it doesn’t. It is very easy to do. A lot of groups have done this.” Unfortunately, Judson continued, this approach does not work particularly well. Even with all the data that are available, the data do not provide enough statistical power to reliably identify relationships between the chemicals and their effects in living organisms. “What you have to do is actually put some biological knowledge into the mix,” he said. “We just don’t have enough data to let statistics drive everything.” One way of inserting biology into the models is called the adverse outcome pathway approach. This involves searching the scientific literature for biochemical pathways involved in the adverse outcomes created by a particular chemical. To describe the approach, Judson offered a hypothetical example involving vascular disruptor chemicals, or VDCs (see Figure 5-1). “We gave this problem to a really good postdoc, telling her, ‘We have some evidence that if you perturb VEGF [which is a particular pathway that is involved in vasculogenesis in a developing embryo], you can cause certain kinds of developmental defects.’ Our smart postdoc then works out, by going to the literature, that VEGF actually disrupts cytoskeletal signaling in endothelial cells. If the chemical does that, then it leads to specific whole-animal defects.” By scanning the literature one can come up with multiple pathways leading from an initial chemical interaction with a pathway to something happening in cells to something happening in tissues to something happening in the whole organism. Environmental chemicals cause toxicity by interacting with specific biological molecules, Judson noted, so one can examine the various biological molecules involved in a particular adverse outcome pathway—for example, VEGF or CCL2 or the aryl hydrocarbon receptor in the case of embryonic vascular disruption—and examine the in vitro databases to see which environmental chemicals affect those biological molecules. “You can now start doing a reasonable job of predicting the kinds of chemicals that we have tested that might have these kinds of phenotypes.” 

APPROAACHES TO PRIO ORITIZING CHEM MICALS 77 FIGURRE 5-1 Adversse outcome patthway approachh. SOURC CE: Knudsen and Kleinstreeuer, 2011. Reeprinted with ppermission froom John Wiley W & Sons, Inc. I Thhere are now w some very y detailed annd complicateed experimennts called “targeted tessting experim ments” being ddone in collabboration withh a numbeer of academiic and industrry groups in aan effort to woork out speciffic pathwaays for speciffic chemicalss, Judson saidd. “This has sshown that you don’t have h to solvee all the probllems for all chhemicals, butt by having thhis big daatabase, you can take speecific chemicaals and undeerstand a lot of detail of what is goiing on all the way to the tooxic endpointt.” It is also possibble to model th he effects of cchemicals onn groups of ceells using what w is called d a “virtual tissue model.”” With such a model one ccan simulaate the behavior of a grou up of cells orr a bit of tissuue over time— — say, thhe tissue in a developing limb bud. Thhe model takees into accouunt the varrious biochem mical pathway ys in the cellls and betweeen the cells thhat controol the develop pment of the limb bud, annd so it becom mes possible to observve how a particular environm mental chemiical affects thaat development. “You canc actually look l ynamics at thee level of cellls and groups of at the dy cells,” Judson said. Byy combining biological b infformation withth in vitro andd in vivo data in this waay, it opens up u a new apprroach to priori ritization, Judsson said. “Onnce you haave those mo odels, then you can take a new chemiical where you 

78 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS don’t have animal data and you can run simple assays and make a prediction. Does this look like it might be a carcinogen? Does it look like it might be a developmental toxicant?” The predictions are not 100 percent accurate, he said, but they are accurate enough to do prioritization. “We are not trying to replace animal tests yet. Maybe one day. But if we want to do prioritization, this approach looks good enough.” In addition to knowing which biochemical pathways are affected by which environmental chemicals, it is also important to know what dose of a particular chemical is necessary to produce an effect. The EPA has developed an approach to creating estimates for what it calls the biological pathway altering dose, which is the amount of a chemical necessary to turn on a particular pathway. “This approach requires two experiments,” Judson explained. “You measure the intrinsic clearance rate of a chemical in liver cells, which could either be a rodent or human, and you measure plasma protein binding. These parameters are then used in a relatively simple computer model that calculates what we call the Css, the Concentration at Steady State per daily dose, which is just the conversion factor: If I take 1 milligram per kilogram per day of a chemical, what is my steady state concentration going to be?” It is also possible to go beyond steady state assumptions. Finally, he said, the in vitro potency is combined with the Css values, and this produces the biological pathway altering dose. This process is all done in vitro, and it costs perhaps $1,000 to get a reference dose for a particular pathway. However, Judson offered an important caveat: “It is still just a concept. We have shown a couple of cases where you get within about an order of magnitude of what you would get from animal studies.” But it is not yet a fully proven approach. Once there is an understanding of the biological effects of chemicals and the dosage at which the effects take place, it is necessary to get information about exposure, Judson said. “Hazard doesn’t mean anything if you don’t have some estimate of exposure,” he said. “You need to have exposure models that are equally high throughput. You have to be able to make some sort of prediction for thousands of chemicals.” To that end the EPA created the ExpoCast3 program for high- throughput modeling of exposure. It is well behind the efforts to model toxicity, he said, but the agency has shown it is possible to make exposure estimates for 10,000 compounds. “Those exposure estimates 3 Further information about ExpoCast is available at http://www.epa.gov/ncct/ expocast (accessed March 31, 2014). 

APPROACHES TO PRIORITIZING CHEMICALS 79 have really wide confidence intervals today,” Judson said. The agency will be making a big effort over the next few years to reduce those confidence intervals, he added, but even with today’s confidence intervals it is possible to use the estimates for prioritizing chemicals for further testing. One of the most important parts of understanding exposure is knowing how a chemical is used, Judson noted. This fact led the EPA to develop the Chemical and Product Categories (CPCat) database, which contains information on about 40,000 compounds.4 Summing up the work on ToxCast, he said that the program has been controversial with some audiences, with some recent presentations and publications claiming that ToxCast has failed. “There are problems,” he acknowledged. “Having said that, though, the EPA is confident that you can use this approach for prioritization. We are not going to ban a chemical or say that a chemical is really bad or really good because of this, but we can start prioritizing chemicals for further testing.” The first real-world application of ToxCast is likely to be the Endocrine Disruptor Screening Program, a congressionally mandated program to put about 5,000 chemicals through a screening process. The problem is that the screen costs about $1 million per chemical, and the worldwide testing capacity is only about 50 to 100 chemicals per year. Thus, with current technology, it would cost $5 billion and take 50 to 100 years to complete. The ToxCast approach should make the screening program feasible. The relevant pathways are known, and the exposure assessment can be done. The EPA has already started, Judson said, and the first outcomes should be available in about 3 years. APPROACHES TO PRIORITY SETTING IN CALIFORNIA Gina Solomon, Deputy Secretary for Science and Health at the California EPA, spoke about how the California EPA is setting priorities on environmental chemicals. The state is not trying to compete with federal programs, such as those at the EPA, she said. Instead her agency is trying to develop complementary programs that can “help move 4 Further information on the CPCat database is available at http://actor.epa.gov/ cpcat/faces/basicInfo.xhtml (accessed March 31, 2014). 

80 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS forward issues around prioritizing chemicals, identifying issues of concern, and taking action when necessary.” There are different approaches to priority setting depending on what one is actually doing, Solomon said—for instance, whether one is interested in doing screening or doing assessments. To illustrate the sorts of priority setting that are taking place in California, she described three statewide programs: CalEnviroScreen, Biomonitoring California, and Safer Consumer Product Regulations, formerly known as Green Chemistry. Each program sets priorities in its own way. CalEnviroScreen Program CalEnviroScreen5 is an environmental justice screening tool, created by the Office of Environmental Health Hazard Assessment, that is used statewide (California EPA and OEHHA, 2013). Version 1.0 was released in spring 2013, and version 2.0 was released in August 2014. Solomon explained that the screening tool contains 17 different indicators, including such things as pesticide use; various air quality indicators; indicators of environmental effects, such as leaking underground storage tanks and toxic cleanup sites; and vulnerability factors, which range from poverty, educational attainment, and linguistic isolation to asthma emergency room visits and low birth weight. The data are mapped to individual zip codes, and they will soon be mapped to individual census tracts. Total pollution scores are calculated from the indicators, with the environmental effects indicators being multiplied by a factor of one-half compared to the human exposure indicators, and the pollution scores are then placed into deciles, said Solomon. These pollution scores are next multiplied by the population vulnerabilities, with the result being a risk score for each zip code in the state. Figure 5-2 is a map of California with the areas in the top 5 percent of risk indicated in blue and the next 5 percent in orange. As can be seen from the figure, many of the highest risk areas of the state lie in the Central Valley, the Imperial Valley, and around Los Angeles. 5 CalEnviroScreen is available at http://oehha.maps.arcgis.com/apps/OnePane/ basicviewer/index.html?appid=1d202d7d9dc84120ba5aac97f8b39c56 (accessed April 4, 2014). 

APPROAACHES TO PRIO ORITIZING CHEM MICALS 81 FIGUR RE 5-2 CalEn nviroScreen 1.11 results: Mappping applicatiion of zip coddes with hiighest CalEnvirroScreen 1.1 sccores. SOURC CE: Californiaa OEHHA, 2014. Reprinted w with permissionn from Californnia OEHHA. What W does thiss sort of environmental juustice screeninng tool have to say ab bout chemicall priority settiing? Solomonn said that ussing such a toool to focu us assessmentts on dispropo ortionately im mpacted comm munities can be importtant in pointting to chem micals that deeserve furtherr scrutiny. FFor 

82 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS example, she said, perchlorate was not identified as an important chemical until it was found in the drinking water supply of the town of Rancho Cordova in California in 1997 after the Department of Public Health developed a new test that had a lower limit of detection. “This was a community that was already known to be impacted from a Superfund site,” she said. “Testing the water supply was a reasonable thing to do in honing down to look at the risks to that community. Lo and behold, this chemical has now become a fairly significant priority nationwide.” Biomonitoring California The second program that Solomon described was the Biomonitoring California program,6 which was established in large part in response to the detection of polybrominated diphenyl ether (PBDE) flame retardants in breast milk. These chemicals were first identified in Sweden in the late 1990s, but soon after the Swedish studies were published, the California Department of Toxic Substances Control laboratory adopted a method for monitoring for the chemicals and reported detecting them in two settings. They were found in harbor seals that had died and washed up in the San Francisco Bay and also in breast biopsy specimens from women in the San Francisco Bay area. The most striking and worrisome part of the finding was that the levels measured in the human samples from around the San Francisco Bay were 40 times higher than the levels from the Swedish study. “This got attention in the media and in the legislature,” Solomon said. “Some of the PBDEs were banned the following year in California.” The push to establish the California Biomonitoring program was the result of the “realization that biomonitoring can be useful for identifying new priorities, new chemicals that we weren’t really thinking about,” she said. There are two aspects to the chemical selection and priority setting in Biomonitoring California. The program starts with the chemicals that the Centers for Disease Control and Prevention (CDC) is looking at. But it also has a scientific guidance panel that can designate and prioritize chemicals that are outside the CDC biomonitoring program list. “We have tended to focus on chemicals that are outside the list,” Solomon 6 Further information on the Biomonitoring California program is available at http://www.biomonitoring.ca.gov (accessed April 2, 2014). 

APPROACHES TO PRIORITIZING CHEMICALS 83 said, “because we feel that the CDC is doing a very excellent job on what they are working on. We want to be value added.” Thus, one of the things that the scientific guidance panel has prioritized is chemicals whose level of use may be different in California than in other states and, particularly, those chemicals whose use may be increasing in California. Thus, the panel has examined chemicals that are serving as replacements for chemicals that have been banned or restricted in California. Among the categories of chemicals that are shifting in California are flame retardants, phthalates and bisphenol A in several uses, and perchloroethylene in dry cleaning, whose use is being phased out across the state. “That has resulted in a number of chemicals—and, more specifically, chemical groups—being listed as priorities for the California Biomonitoring program,” Solomon said. Thus, the scientific guidance panel designated several categories of flame retardants as a group even though they are somewhat different chemicals structurally, because they are used in similar niches. “There is going to have to be additional prioritization within those,” she added, “but our laboratories are working on developing methods to screen for as many of those as they can.” The panel is also looking at a broad collection of various bisphenols, some of which are potential replacements for bisphenol A and others of which have somewhat different uses. The various chemicals were grouped together into a chemical class, Solomon said. There are some categories of chemicals that the panel decided not to prioritize, she said, such as synthetic hormones used in food production and antimicrobials used in food production. The reasons for not prioritizing the various categories included because the chemicals in a category were too disparate, because the chemicals were too difficult to biomonitor effectively, or because there was no clear California-specific concern related to the chemicals. Most recently, Biomonitoring California is preparing to do non- targeted testing with a time-of-flight mass spectrophotometer, which allows screening for unknown chemicals—those that are not specifically included in current biomonitoring assays—in environmental or biological samples. Nontargeted screening can help identify potential exposures and set priorities for testing, risk assessment, or ultimately mitigation. “If you sample a whole lot of people and start consistently seeing something that you weren’t expecting to see,” Solomon said, “it is time to develop a method and start specifically looking for that in populations of interest, 

84 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS evaluating the risk, and deciding if further action is needed. It is definitely a very useful tool for priority setting.” California Safer Consumer Product Regulations The third program is California Safer Consumer Product Regulations7 under the Department of Toxic Substances Control. The initial chemical priority setting in that program largely relies on lists from others. The program currently uses 23 lists containing approximately 1,200 chemicals or chemical groups, she said. “This is more of a risk management kind of program,” she explained. “We are not trying to reinvent the chemical priority-setting wheel, here. We are trying to focus on prioritizing at the product level, getting companies to look at alternatives and ultimately to move gently towards safer alternatives.” From this initial list of about 1,200 candidate chemicals, the program is focusing on about 200 chemicals for the initial round of product selection, and based on these 200 it will choose a set of up to five priority consumer products. For each of these priority products there must be potential exposure to the candidate chemicals and also the potential for the exposures to contribute to or to cause significant or widespread adverse impacts. Looking for a chemical in a product is especially tricky, Solomon said, because there is so little information available on which chemicals are being used where. “I have lots of examples of people in our agency who are phoning up companies saying, ‘I want to get a quote on this. Can you tell me what products you use? Can you send me the names of what you are actually selling?’ Or else they are going to stores and looking at what is on the shelf. We are just scrambling to figure out what is used in California in what amounts.” The initial list of five priority products will be just the beginning, Solomon said. After that list is released, the program will be putting out a work plan and developing a much more ambitious list. ASSESSING AND PRIORITIZING RISKS IN CANADA The next presenter was Heather Patterson, Senior Evaluator in the Healthy Environments and Consumer Safety Branch of Health Canada, 7 Further information on California Safer Consumer Product Regulations is available at http://www.dtsc.ca.gov/SCP (accessed April 2, 2014). 

APPROACHES TO PRIORITIZING CHEMICALS 85 which is Canada’s equivalent of the U.S. Department of Health and Human Services. She is currently working on developing innovative approaches for prioritization and assessment, and she described efforts in Canada to assess and prioritize chemicals. The legislation underpinning Canada’s efforts in that area, Patterson explained, is the Canadian Environmental Protection Act, which was passed in 1988 and amended in 1999. It provides the regulatory framework in Canada for information collection, risk assessment, and risk management of new and existing chemicals and organisms. It includes provisions for the assessment of existing chemicals, she said, and it requires that every new substance made in Canada or imported into Canada be assessed against specific criteria. According to the act, the Minister of the Environment must maintain an inventory of existing substances in Canada, known as the Domestic Substances List (DSL), and this list is the sole basis for determining whether a substance is deemed to be new or existing in Canada. The 1999 amendments required the ministers of the environment and health to categorize the approximately 23,000 substances that were on the DSL, using specific criteria to identify priorities for future assessment work. Patterson noted that Environment Canada looked to see which of those 23,000 substances had the potential to be either persistent or bioaccumulative and which of those were inherently toxic to non- human organisms, while Health Canada looked to see which of the substances posed the greatest potential for human exposure and which ones were likely inherently toxic to humans. In Health Canada’s work determining which substances had the greatest potential for human exposure, it focused on three lines of evidence, Patterson said. It looked at the amount of a substance in commerce, the number of identified companies manufacturing or importing a given substance, and the use codes or the uses of the substances that were identified. “This was based on the data that was given to us from 1984 to 1986,” she added. “It was old even at the time of categorization, but it was the only information that we had for every substance on the DSL so we could compare equally across all of them.” To identify which substances for further assessment were potentially toxic to humans, Patterson said, the agency performed a “list-matching exercise,” looking to see which substances had been classified by other agencies as carcinogens, mutagens, or reproductive toxicants. When Health Canada combined its priorities with those identified by Environment 

86 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS Canada, the results identified 4,300 substances for future assessment work. As this categorization work was being wrapped up, Canada’s Chemicals Management Plan was announced.8 That plan, which is the government of Canada’s response to the Strategic Approach to International Chemicals Management, is designed to meet the 2020 goals set by the World Summit on Sustainable Development for the sound management of chemicals. She noted that it provides a framework for assessment and for the management of the priorities identified through categorization, and it integrates multiple federal programs into a single strategy to ensure that the chemicals are managed appropriately in order to prevent harm to Canadians and their environment. The Chemicals Management Plan has three phases: 2006–2011, 2011–2016, and 2016–2020. In the first phase, she said, the highest- priority substances were addressed and work was initiated on the lower- priority substances. The major focus of the second phase, which is now ongoing, is the substance groupings initiative. In the third phase the remainder of the 4,300 priorities will be addressed. Patterson explained that the highest-priority substances dealt with in the first phase of the plan were about 500 substances that met the criteria of potential persistence, bioaccumulation, and inherent toxicity to aquatic organisms that had been identified by Environment Canada, or that had high exposure potential, and that were identified as posing a high hazard for human health. These substances were addressed through three different mechanisms. There were about 150 substances that were persistent, bioaccumulative, and inherently toxic but that were believed to no longer be in commerce in Canada, Patterson said. “For those substances, we published SNAcs. This is a Significant New Activity. It is very similar to a SNUR [significant new use rule] in the U.S., which means that if anyone wants to use these substances for a new activity, they need to notify us before they can do so.” There were about 200 substances that were of high concern and were found to be in commerce, she said. These were assessed under the Challenge Program, which consisted of a variety of individual screening- level assessments of the substances. 8 Further information on the Chemicals Management Plan is available at http://www.chemicalsubstanceschimiques.gc.ca/plan/index-eng.php (accessed April 2, 2014). 

APPROAACHES TO PRIO ORITIZING CHEM MICALS 87 Finnally, there were w about 16 60 substancess of high conncern that weere found to be used predominantly p y within the ppetroleum secctor. A focussed sectoriial approach was develop ped to deal with these ssubstances thhat relied on exposure-based prioritiization (see F igure 5-3). Deepending on a substance’s characterisstics—such aas whether tthe substance left the faacility where it was produ ced and whetther it was ussed by thee public, only y by industry,, or by other sectors—thee substance w was classiffied as being in Stream 0, Stream 1, Strream 2, Streaam 3, or Streaam 4 (Streeam 0 was used u for produ ucts that werre similar in composition to petroleeum substances, but weere not mannufactured orr used by tthe petroleeum sector). “The “ assessmment approachhes that we ussed to deal wiith these substances were w tailored d according to the expoosure scenariios identiffied for each stream,” s Patteerson said. In addition to its work on o the high-ppriority subsstances, Heallth Canad da was also id dentifying low w-priority subbstances. Thiis is a differeent sort off exposure-bassed prioritizatiion method, P Patterson notedd. In identifyiing low-prriority substan nces, the preemise was “nno exposure eequals no riskk,” she saiid. “We wanteed to focus on n this early inn the Chemicaals Managemeent Plan because b it allo ows us to iden ntify substancces with low concern so thhat we can n focus our resources on thoset substannces with highher concern.” It turnedd out that a nu umber of sub bstances that had been ideentified throuugh categoorization as priiorities actuallly had a low ppotential for exxposure becauuse FIGUR RE 5-3 Petroleeum-sector streeam approach. SOURC CE: Patterson, 2013. 

88 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS very low volumes of the substances were manufactured or used in Canada. “Environment Canada addressed these substances by doing worst-case modeling to calculate predicted environmental concentrations and compared those with toxicity values,” she said. “At Health Canada, we did not quantify exposure, but rather we identified which substances had no or very low exposure potential.” Exposure could be direct or indirect. By their nature, those substances that were used in very low volumes had low potential for indirect exposure via the environment. Thus, the main issue was the potential for direct exposure. In particular, was a substance used in consumer products? “For substances that are used in consumer products,” she said, “even if there is only a low volume in commerce, there could still be a high potential for exposure if the substance is applied directly to your skin, for example.” For those substances found to not have the potential for indirect exposure and to not be used in consumer products, Health Canada concluded that further assessment work was not necessary. After carrying out three assessments on a total of approximately 1,200 substances, which were published in 2013, the agency found that approximately 700 of the substances needed no further assessment work. The other 500 will be considered further in future assessments, Patterson said. “We expect that we will be using this approach again to deal with as many substances as possible once we collect current commercial status on the remaining priorities.” As the first phase of the Chemicals Management Plan was being wrapped up, those who were responsible for its implementation began working on priorities for the second phase, Patterson said, and what became obvious was that doing assessments of single individual substances was not always the most efficient approach. So they developed an approach by which they would collect similar substances into groups and use those groupings in their prioritization and assessment work. The groupings were developed based on a number of different factors: common chemical classes, common modes of action, common uses, common sectors, and so on. “We also considered scheduling implications,” she said. “We wanted to look for the availability and timing of international information. For example, if we are expecting a major assessment report from the United States, we would like to time our assessment report to either be closely related with that or maybe even follow it to allow for consideration of all possible information.” They also considered the ability of various stakeholders to participate. “We 

APPROACHES TO PRIORITIZING CHEMICALS 89 tried not to align all of our metals groupings, for example, at the exact same time because that sector would be overwhelmed with having to provide data and comments.” The result was the Substance Groupings Initiative, which covers about 500 substances divided into nine groups, such as aromatic azo- and benzidine-based substances (which is by far the biggest grouping, with more than 300 substances), cobalt-containing substances, certain organic flame retardants, and phthalates. One grouping, labeled Certain Internationally Classified Substances, is not actually a group of similar substances; rather it is a group of individual substances that the planners felt warranted attention due to their international high hazard classifications. Pulling back and looking at all 4,300 substances requiring assessment as a result of categorization, Patterson offered an overview of the different assessment approaches being used (see Figure 5-4). As described above, there were 500 highest-priority substances that were divided into SNAcs, Challenge Substances, and petroleum-sector substances; 700 that needed no further assessment because there was low potential for exposure to them; and 500 in the Substances Groupings Initiative. She says there are likely to be another 200 or so substances that will fit into the petroleum-sector stream approach and another 700 substances or so that will be found to have low exposure potential after further data collection. “We also have about 700 polymer substances that we are dealing with at this time,” she said. “We are developing an approach to deal specifically with those substances.” That leaves about 1,000 more substances for which the assessment approach has not yet been determined. Now that Health Canada is moving toward the middle of the second phase of the plan, it is time to think about how to approach the third phase, Patterson said. “It is not so much about selection anymore,” she said. “We have to figure out what the best way is to deal with what is left.” To do that she and her colleagues are again looking at how to group these substances based on such things as structure, mode of action, functional use, and possible substitutions. They are also looking at what the potential exposures are for each of the substances. One important issue is the commercial status in Canada for each substance. How much of it is in commerce, and how is it being used? Is it used in consumer products? And has it been observed in environmental media or biomonitoring studies? 

90 ENVIRONMENTAAL HEALTH RISK KS OF CHEMICA ALS FIGUR RE 5-4 Assessm ment approach hes for categoriization prioritiees. NOTE: PBiT = persistent, bioacccumulative, annd inherently toxic, SNAcs = significcant new activiities, TBD = to o be determinedd. SOURC CE: Patterson, 2013. Annother issue is the availabbility of toxiccity data. Is there empiriccal data on a given sub bstance? Howw much? Is iit positive orr is it negativve? What end points arre involved? Is there moree than a meddian lethal doose (LD50)9 available? Are there high-throughpuut screening data availablle? As an example, shee described th he results of a TOXLINE search for daata on thee 2,300 remaaining priorityy substances . Fifty-three percent of tthe substances had no hits h at all, andd 18 percent hhad very few w hits, and sinnce many of the hits forf those 18 percent are llikely to not be relevant to human n health, therre may be only o 29 perceent of the 2,,300 remaininng priority substances that have emmpirical data rrelevant to thee assessment of human n health. “I kn nly one sourcce,” she said. “There may be now this is on 9 Median lethal dosse (LD50) is the t statisticallly derived meedian dose off a chemiccal or physical agent expecteed to kill 50 peercent of organanisms in a givven populattion under a deefined set of co onditions (IUPA AC, 2007). 

APPROACHES TO PRIORITIZING CHEMICALS 91 other data out there and we are still gathering data. But this really highlights the need for innovative assessment approaches.” Finally, there are scheduling issues. These include the identification of possible opportunities for international collaboration and the alignment with data generation. For example, she said, Richard Judson is “coming up with some great new high-throughput ideas.” She noted that we need to wait until we get those data and determine how best to use them before beginning to assess substances for which that is the only source of toxicological data. In carrying out assessments, it is important to identify emerging priorities, Patterson said. It is well recognized, for instance, that the scientific understanding of exposure and toxicity continues to evolve over time and that global regulatory action on chemicals also changes over time. “So we can’t just stop with using the categorization decisions to decide on priorities,” she said. “We have to continue to update our list of priorities based on the evolving landscape.” Traditionally, she continued, there are seven “feeders” that are used for the identification of priorities: categorization decisions, industry information, decisions from other jurisdictions in Canada, international assessments or data collection, public nominations, trends in new substance notifications, and emerging science or monitoring data. “We are currently looking for a way to make this process more systematic so that we can ensure we are looking at all of the appropriate pieces of information in a timely manner,” she said. To conclude, Patterson offered a list of lessons learned to date: • There are many limitations to conducting a priority-setting exercise that is based on dated inventory data, but it is often difficult to get new data. • There is a lack of approaches available for modeling substances other than the generic organics. “When it comes to the inorganics, UVCBs [Unknown or Variable compositions, Complex reaction products and Biological materials], and polymers, for example, if we don’t have empirical data, there is not a lot we can do.” • Indirect exposures, such as through environmental media, do not typically drive human health assessment outcomes. • Instead, direct exposures—i.e., consumer product exposures— are more typically the key drivers in assessment outcomes. “We typically have to use upper-bounding models to develop scenarios for these substances. We refine them if the data are 

92 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS available. What we have found is that it is often not easy to obtain data to refine these scenarios.” • A substance-by-substance approach is less efficient for both risk assessment and risk management. • However, assessment of substance groupings can also be quite challenging. Groups built for one purpose are not always well suited for others. “If we build a grouping based on informed substitution, the risk assessment is often quite challenging because those substances could have very different exposure patterns or health implications.” In the discussion period that followed all of the session’s presentations, Liz Harriman with the Massachusetts Toxics Use Reduction Institute asked Patterson about the benefits of grouping as well as any cautions. She answered that there are pros and cons to grouping. “Obviously, for the substances that have absolutely no data, building a group can allow read across from a data-rich substance to the data-poor substances within that group,” she said. “That is how we are dealing with a lot of substances right now that have no empirical toxicity data available on them.” On the other hand, she said, building a group for one purpose often makes a grouping quite detrimental for other purposes. “Building a group based on a common mode of action may make it almost impossible to do risk management in the end or vice versa,” she said, “while building a group based on a sector makes it really difficult to do a risk assessment. Substances used in the same sector can still be used in very different ways.” AMERICAN CHEMISTRY COUNCIL VIEWS ON CHEMICAL PRIORITIZATION The final speaker was Christina Franz, Senior Director of Regulatory and Technical Affairs at the ACC. She described the ACC’s views on chemical prioritization processes and on how to improve the prioritization process used by the EPA. She began with some background on how the ACC came to develop a prioritization tool and the purpose of that tool. In 2009 the ACC published 10 principles for modernizing the Toxic Substances Control Act (TSCA) of 1976.10 One of those principles called for the EPA to systematically 10 Toxic Substances Control Act of 1976, Public Law 94-469, 94th Congress. 

APPROACHES TO PRIORITIZING CHEMICALS 93 prioritize chemicals for safe use determinations. Coincidentally, during that same year the EPA also published Principles for TSCA Modernization, which called for manufacturers and the EPA to assess and act on priority chemicals. Since that time Congress has held a number of hearings on bills proposed to modernize TSCA, and, Franz said, there has been a general consensus among the witnesses at these hearings that prioritization is essential. In 2012, the EPA came out with a list of 83 priority “work plan” substances that it announced it would be doing targeted risk assess-ments on from 2012 through 2015. And the Chemical Safety Improvement Act of 201311 became the first bill introduced in Congress to modernize TSCA that has included a prioritization section within the statute. The ACC agrees with this emerging consensus on prioritization, Franz said. “It is clear that, from a practical standpoint, we have limited resources and limited time, and this requires that we focus on those substances that are of highest priority for further evaluation,” she said. Thus, in particular, the ACC sees the EPA’s work plan for a chemical prioritization process as an extremely important step forward for the agency. TSCA does not specifically direct the EPA to undertake such a prioritization, but it was within the agency’s authority to do so. When the EPA first published its proposed priority-setting approach, that approach focused exclusively on hazard, Franz said, and it ignored the exposure part of the prioritization equation. The ACC was pleased that, after listening to various stakeholders, the EPA decided to integrate hazard and exposure factors to identify substances for further evaluation. Still, she said, the ACC believes that the EPA’s process can be improved—not for the 83 substances that have already been identified for targeted risk assessments, but rather in moving beyond those to a broader set of priorities for further evaluation. “The process that the agency did employ was not sufficiently based on objective, science-based criteria that could be applied consistently across all chemicals evaluated,” Franz said. “They began their prioritization assessment by looking at lists of chemicals that currently existed and then whether substances within those lists met certain factors that the agency was concerned about. The result, in ACC’s view, is that there are inherent biases that exist. The priorities identified might not actually represent the highest hazard and greatest exposure potential and therefore could be a waste of time and resources on the part of the Agency.” 11 Chemical Safety Improvement Act of 2013, S. 1009, 113th Congress, 1st session (May 22, 2013). 

94 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS In short, she said, prioritization should integrate hazard and exposure criteria. “If you have high exposure with no hazard, it shouldn’t be a concern. Similarly, high hazard with no exposure, also not a concern. The highest priorities really exist at the intersection of highest hazard and greatest exposure.” In 2009 the ACC developed a prioritization tool that would embody these principles,12 said Franz. It refined the tool in 2011 so that the tool could deal with chemicals lacking sufficient information for priority setting, would be better aligned with the Globally Harmonized System of Classification and Labeling of Chemicals (GHS), would incorporate scientific advances regarding persistence and in bioaccumulation, and would have increased scientific rigor (ACC, 2011a). The concept behind the tool is of a matrix with hazard on the vertical axis and exposure on the horizontal (see Figure 5-5). The hazard level is classified as low, medium, medium high, or high, using the same criteria used by the GHS. The hazard is scored separately for human hazard and environmental hazard, and the final hazard ranking is based on the higher of the two. If existing sources of information on a chemical provide insufficient information to determine the hazard level, the highest hazard score is assigned. The exposure ranking for a chemical is determined by adding scores from three components: the use pattern, production volume, and persistence and bioaccumulation. The use pattern score is derived from the Chemical Data Reporting Rule, the rule by which, under TSCA, companies report periodically on substances that are currently in commerce. Substances used by consumers get a score of 4, those in commercial use get a 3, those in industrial use get a 2, and those in intermediate use get a 1. Similarly, production volume scores are 4 (more than 100 million pounds), 3 (1 million to 100 million pounds), 2 (25,000 to 1 million pounds), and 1 (less than 25,000 pounds). The persistence and bioaccumulation scores are 5 for substances that are both persistent and bioaccumulative, 3 for those that are one but not the other, and 1 for those that are neither persistent nor bioaccumulative. The three scores are added together to get the total exposure score, which is then used to assign substances to an exposure band or range: low (3 or 4), medium low (5 or 6), medium (7 or 8), medium high (9 or 10), and high (11 or 13). 12 Further information on ACC’s approach to prioritization is available at http://www.americanchemistry.com/TSCA (accessed April 2, 2014). 

APPROAACHES TO PRIO ORITIZING CHEM MICALS 95 FIGUR RE 5-5 Concep pt behind priorritization tool. CE: ACC, 2011b. Reprinted with permissioon from the AC SOURC CC. Thhe exposure ranking r for a chemical is ddetermined byy adding scorres from three compo onents: the use pattern,, productionn volume, annd persisttence and bio oaccumulation n. The use paattern score iis derived froom the Ch hemical Dataa Reporting Rule, R the rulle by which,, under TSCA, compaanies report peeriodically on substances thhat are currentlly in commercce. Substaances used byy consumers getg a score off 4, those in ccommercial uuse get a 3, 3 those in inddustrial use geet a 2, and thoose in intermediate use get a 1. Sim milarly, produ uction volum me scores are 4 (more thaan 100 million pounds), 3 (1 milllion to 100 million m poundds), 2 (25,0000 to 1 million pounds), and 1 (less than 25,000 2 pounnds). The ppersistence annd bioacccumulation sccores are 5 fo or substances that are bothh persistent annd bioacccumulative, 3 for those th hat are one bbut not the other, and 1 ffor those that t are neitheer persistent nor n bioaccum mulative. The tthree scores aare added together to get the total exposure sccore, which iis then used to assign substances tot an exposurre band or raange: low (3 or 4), mediuum low (5 or 6), mediumm (7 or 8), meedium high (99 or 10), and hhigh (11 or 133). Suubstances are then placed on the priorit itization matrrix according to their hazard h and exposure e ores, with a total prioritiization ranking sco calculaated by addiing up the hazard h rankiing (1 througgh 4) and tthe exposu ure ranking (1 through 5). The total priioritization raanking is thuss a numbeer between 2 and 9, from lowest hazarrd/lowest expposure (Prioriity Groupp 2) to higheest hazard/hig ghest exposuure (Priority Group 9). T The 

96 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS intention, Franz said, is that an agency should focus on substances in Priority Group 9 first as being the highest priority for further evaluation. There is also a second tier to the prioritization process that allows for more qualitative scientific judgment, she added. The qualitative factors can be used to move a substance up or down in priority within a given priority group and thus create a rank order within a priority group. The factors considered here are biomonitoring, whether a substance is used in a children’s product, emissions information, and whether there are any international risk management actions pending on a particular substance. “If you are in Priority Grouping 9, any one or more of these considerations would, perhaps, move you to number one in Priority Group 9,” she said. In conclusion, Franz offered a number of benefits that using the ACC prioritization tool provides: • The prioritization tool is based on objective scientific criteria regarding both hazard and exposure. • It addresses human health and environmental safety. • It is transparent and offers a helpful matrix visualization. • The highest priorities are very clear. • The exposure indicators used are use, volume, and persistence/ bioaccumulation. The final indicator incorporates scientific advances regarding persistence and bioaccumulation. • The tool is flexible and can be updated to accommodate improved scientific information. • The qualitative factors can be used to influence rank ordering within a priority grouping. DISCUSSION Lynn Goldman asked a general question of the panel members whether they heard things from each other that they could apply in the own areas, particularly concerning the issue of prioritization. Richard Judson replied that his group, which consists mainly of chemists and biologists, started from a hazard-centric standpoint, but it has spent a great deal of time over the previous 6 to 8 years talking with the Canadian group. Betty Meek of that group had recommended that instead of initially focusing on hazard, they start on the exposure side. “We felt that our high-throughput hazard approach would let us set priorities first, 

APPROACHES TO PRIORITIZING CHEMICALS 97 which we could then refine with lower-throughput exposure estimates,” Judson said, “but now we have found ways to do the initial exposure estimates computationally for most chemicals, while we can’t currently run the hazard screens for the complete set of tens of thousands of chemicals. There is some low hanging fruit on the exposure side.” The lesson he has learned is that it is important to move back and forth between different viewpoints. “You do as much exposure as you can, using inexpensive models. Then do as much high-throughput hazard measurement as possible. You go back and forth, which is a little bit like what the ACC tool is doing. You are almost prioritizing prioritization.” Gina Solomon answered that the different approaches are not mutually exclusive and that having multiple approaches to determining priorities can be beneficial because the priorities being set might depend on context—not just the regulatory, but also whether the prioritization is at a community or statewide or national level and whether one is prioritizing for screening or testing or risk assessment or risk management. “I like multiple approaches to priority setting,” she said. “Let’s do several different things and hope that we can then elevate the things that should be elevated.” Luz Claudio, from Mount Sinai School of Medicine and a Roundtable member, commented that many in vitro models have been around for more than 20 years and asked Richard Judson, “Are we any closer to a clear guideline of how to use in vitro testing—whether for screening or just to make a dent in the big list of chemicals that have not even been tested in any way?” Judson began by noting that there are old in vitro tests and new in vitro tests. “Most of what we are using was developed in the last 5 to 10 years, and there are new ones becoming available all the time.” These new tests are more focused on molecular targets—molecules that are known, when triggered, to lead to certain types of phenotypic changes. “Then there are intermediate phenotypes,” he continued. “Chemicals cause whole-cell changes. Many of the new systems have multiple cell types in the same well, which allows us to see the effects of cell-cell communication. It is getting more and more complex. For instance, some of the new systems have xenobiotic metabolism active within the well.” And there is a lot of new technology coming online, he added. “By running lots of assays and combining them computationally, we believe we are getting a better picture of what would go on in a whole animal.” Judson then noted that a major issue in the regulatory area is to validate such tests. One of the reasons that so many old tests are still 

98 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS being used, he said, is that even for simple tests it can take 5–10 years to go through the validation process. “There are other realms of medical testing where that would be crazy, but, somehow, the toxicity testing area has allowed this approach that, oh, a test has to be perfect. . . . That has made it difficult to even imagine how we would move these new tests into the regulatory area.” His group has suggested the formation of an international group that could come up with ways to speed up the validation process, he said. Until that happens, there will be great uncertainty concerning just when the new tests will start being used. Dennis Devlin, from ExxonMobil Corporation and a Roundtable member, asked Judson about the ToxCast project. “I think many of us, if not all of us, were disappointed that ToxCast didn’t show more predictive power, at least currently with the information that we have,” Devlin said. Then he asked Judson if he thought that these in vitro assays could be used credibly for questions that need relatively immediate answers. Devlin had heard, for example, that the ToxCast assays had been used to screen the dispersants that had been used in the Deepwater Horizon spill, and he wondered how that could have been done credibly. Devlin had also heard it suggested that ToxCast assays could be used to screen the hydraulic fracturing fluid used in wells. Again, he wondered if that could be done credibly, given that there are thousands of sites and every well will have a different mixture. In such cases, Devlin noted, it is assumed that there will have been some exposure, so he asked Judson, “Do you envision a time where the assays will be able to do that for just a hazard assessment?” Judson noted that there have been two or three talks and papers that have said ToxCast failed. “We are a big science program. We are a target for people.” Still, he said, it is important for him and his colleagues to understand the criticism and to take it seriously. There are three ways in which the program might fail, Judson said: The data could be wrong, there aren’t enough assays or the right assays, or the models are not sufficient. The data are certainly not perfect, Judson acknowledged. “Most of the critiques are saying, you have failed because you can’t exactly replicate in vivo rodent animal studies,” he said. One of the things that has not been included in the work is pharmacokinetics data. “We are just now getting enough of that data,” he said. “If you don’t have pharmacokinetics, that is a huge driver of a chemical’s potential to be toxic. You really have to get the dosing right. We have started to show 

APPROACHES TO PRIORITIZING CHEMICALS 99 that if you put the toxicokinetics there, you improve the estimates. And finally, we know that we don’t have all the right pathways assayed.” The more important issue, Judson said, is that the biology is really complicated. Effects are not at all linear. “By including more biology, we will get there,” he said. “That is not today or tomorrow. I think we will get there just because I believe that biology is not magic.” As for the specific issue of the use of ToxCast for screening, Judson explained that his group had been asked a very specific question concerning the Deepwater Horizon spill. “There were seven or eight dispersants,” he said. The exact ingredients in them were confidential business information, but it was believed that some of them contained nonylphenol ethoxylates, which will degrade to nonylphenol, which is an estrogenic. The worry was that if a large amount of estrogenic material was deposited into the shrimp breeding grounds along the Gulf Coast, there would be a population crash. “The manufacturers wouldn’t publicly say what was in there,” Judson said. “The idea was, we will just take some of those dispersants and test whether they are estrogenic or not. We did that. It turned out there were two [substances] in which we got a small signal. The one that was approved to be used didn’t [have a signal].” Later, he said, through sleuthing through the Internet his group found out that the analysis did get the right answer. Thus, ToxCast was successfully used to answer a specific question. As for fracking, Judson said, his group has not been asked to do such an analysis yet, but they could. “A similar experiment we have been asked to do is take river water samples downstream of an effluent, which are complicated mixtures, and ask, is there anything that looks estrogenic or androgenic or that is hitting the aryl hydrocarbon receptor?” With their high-throughput screens they have been able to successfully do the same sorts of analyses that have traditionally been done by more complicated and more expensive in vitro screening. “We can answer that sort of question,” he said. REFERENCES ACC (American Chemistry Council). 2011a. ACC prioritization screening approach. Available at http://www.americanchemistry.com/Prioritization- Document (accessed April 2, 2014). 

100 ENVIRONMENTAL HEALTH RISKS OF CHEMICALS ACC. 2011b. PowerPoint presentation on ACC’s prioritization system. Available at http://www.americanchemistry.com/Policy/Chemical-Safety/ TSCA/PowerPoint-Presentation-on-ACCs-Prioritization-System.pdf (accessed January 28, 2014). California EPA and OEHHA (California Environmental Protection Agency and Office of Environmental Health Hazard Assessment). 2013. California communities environmental health screening tool, version 1.1 (Cal- EnviroScreen 1.1): Guidance and screening tool. Available at http://oehha. ca.gov/ej/pdf/CalEnviroscreenVer11report.pdf (accessed April 2, 2014). California OEHHA. 2014. CalEnviroScreen 1.1 results: Highest scoring ZIP codes with CalEnvironScreen 1.1 scores. Available at http://oehha.maps.arcgis. com/apps/OnePane/basicviewer/index.html?appid=5e1542837d4246b282dd baa92b0e790f (accessed April 4, 2014). IUPAC (International Union of Pure and Applied Chemistry). 2007. Glossary of terms used in Toxicology, 2nd edition. Pure and Applied Chemistry 79(7):1153–1344. Available at http://sis.nlm.nih.gov/enviro/iupacglossary/ glossarym.html (accessed April 2, 2014). Knudsen, T. B., and N. C. Kleinstreuer. 2011. Disruption of embryonic vascular development in predictive toxicology. Birth Defects Research, Part C: Embryo Today 93(4):312–323. Patterson. H. 2013. Health Canada’s experience with existing substances under the Canadian Environmental Protection Act. Presentation at the Institute of Medicine Workshop on the Identifying and Reducing Environmental Health Risks of Chemicals in Our Society, Washington, DC. Zang, Q., D. M. Rotroff, and R. S. Judson. 2013. Binary classification of a large collection of environmental chemicals from estrogen receptor assays by quantitative structure-activity relationship and machine learning methods. Journal of Chemical Information and Modeling 52(12):3244–3261.