The Endocrine Disruptor Screening program, besides looking at endocrine activity, were also needing to characterize potential exposure to these chemicals for a variety reasons and we can use that information to sharpen our prioritization of chemicals going into further screening and testing. So in this section, I'll talk about Exposure-Based Chemical Prioritization. In recent years, the EPA has been developing high-throughput models for exposure-based chemical prioritization. We're using those in a project that is called ExpoCast or the ability to forecast exposure of chemicals across thousands of chemicals for humans and wildlife and I refer you to a publication in 2013 in Environmental Science and Technology, there might be some more recent papers coming from this group as well, introducing ExpoCast and some of these high- throughput alternatives or approaches to modeling chemical exposure. For the Endocrine Disruptors Screening Program, we've been held to make use of some of these early results and start to consider ways to incorporate this information into our Endocrine Screening Program. We looked at the results for over 7,000 almost 8,000 chemicals these exposure predictions, examine chemicals targeted by the Centers for Disease Control biomonitored in humans as a way to ground truth or compare the prediction results to biomonitoring results looking at results in blood and urine samples to develop chemical descriptors and use the information to predict exposures for these monitored chemicals as well as unmonitored chemicals found in a wide range of consumer products other uses and to identify chemicals that were in direct use, if you will, in consumer products, pharmaceuticals, pesticides, as opposed to indirect use, near-field exposure versus farfield exposure, near-field being things they would take into your home residential use, far field being things that would be a more general environmental exposure and then using this high-throughput model, the researchers extrapolate the predictions to chemicals without biomonitoring data after ground truthing or validating the model against those with monitoring data from the CDC. You can see in this diagram an indication of the results for these almost 8000 chemicals. They're ranked by the one sided upper 95th percentile credible limit in the total demographic, so there is a total demographic or predicted population or predicted population distribution for either a total population or just a subset of predicted 6 to 11 year olds, that's indicated in blue for total and then red for the 6 to 11 and see generally a little higher predicted exposure level for the smallest age group, the 6 to 11 year olds. And each of these results in the X-axis is indicating an individual chemical. So you can see chemicals at the left are predicted to be the highest level of exposure, whether you're looking at the median prediction or the upper 95th percentile and you can see distribution of exposures ranging across orders of magnitude. But you can also get some indication of the uncertainty ranging across orders of magnitude for these predictions of exposure. You can see that the enhanced chemicals. These are the CDC biomonitoring data include a few of the highest predicted exposure chemicals but also a lot of chemicals that are expected to be much lower and are measured to be at a much lower level of exposure. So one takeaway from this diagram might be that some of these other chemicals in the left-hand portion of the panel that are predicted to be a higher level of exposure and perhaps should be a larger focus for the CDC NHANES program process are the considerations to take into account as well. So these first high-throughput exposure predictive models could be very useful for something like the endocrine disruptor screening program where we're looking to prioritize chemicals particularly nonpesticide and chemicals for ongoing screening and testing. So to do this, we need to make some comparison between exposure prediction and in vivo bioactivity. To do this, we need to make a connection between in vitro to in vivo bioactivity. So many cases, we're measuring bioactivity in vitro but we want to extrapolate that to in vivo levels. To do that we need to understand the kinetics, the symmetry of the chemical not in vitro only but in vivo. So we need to understand what it would take in terms of an exposure to reach blood concentrations or tissue concentrations similar to what we're running in the in vitro assays, the high-throughput screening assays. So we have to account for how the chemical gets into the animal or the human and how it's processed or metabolized. To do this, we come up with an approach at EPA of using a number of measurements and putting it together to predict the reverse toxicokinetics, if you will, of how to go from a certain blood concentration to what an exposure level would have to be to reach that blood concentration whether it was in humans, rats, or fish. To do this we need to generate a few pieces of data. One is hepatic clearance data. So we're using cryopreserved hepatocytes with either humans, rats, or fish to run chemicals through and remove aliquots at various time points and to do analytical chemistry to predict at which point, what the half life if you will for that chemical is in response to a certain species hepatocytes, the hepatic clearance rate that along with plasma protein binding and some other physical properties of the chemical allow us to do a pretty good job of predicting what oral exposure would be necessary to reach the blood concentration where we're seeing activity for a chemical endocrine activity in vitro. So we call this the in vitro to in vivo extrapolation or IVIVE. And this is the IVIVE or extrapolation from AC50, so in vitro activity concentrations, to oral equivalents. So this is the bioactive in vitro concentration converted into an estimated steady state oral equivalent in vivo dose. So it's allowing discrimination of chemical potencies. This top plot is a range of different compounds, several hundred compounds left to right and their relative bioactivity across a range of different assays and you can see a lot of overlap. It's pretty much a straight line across. If you do the in vitro to in vivo extrapolation if you had not only this dynamic data but the kinetics data how these chemicals are processed in vivo to give you different concentrations, you start to see much greater degree of separation or discrimination between these chemical potencies. So adding the metabolic or dosimetry component, the kinetics if you will, to the dynamics data allows us to discriminate chemical potencies in much greater and if only have the toxicodynamic or the activity data. So the method for the endocrine disruptive screening program is to bring together the bioactivity data that we're getting from ToxCast and Tox21, the high-throughput screening data with the exposure data we're getting from ExpoCast. Doing an in vitro to in vivo extrapolation in the HTTK, high-throughput Toxicokinetics. I apologize. Another acronym for the IVF approach to integrate bioactivity and exposure and allow us to rank chemicals based on that bioactivity exposure ratio. So we need to extrapolate from in vitro to in vivo using high-throughput toxicokinetics in vitro bioactivity and compare that oral equivalent exposure that would give us bioactivity to the predicted exposures for these thousands of chemicals and to use that to prioritize chemicals for further screening and testing because it's only chemicals where we have exposures approaching where we would expect to see estrogen androgen or thyroid bioactivity that we have a priority for further screening and testing of those chemicals. So let me show you some preliminary results from sort of the first pass of this approach. This is the Integrated bioactivity exposure ranking for based on estrogen receptor bioactivity. So this is a plot data for about 50 chemicals. They're individually listed at bottom on the X-axis and in the Y is the estrogen receptor oral equivalent dose over the predicted exposure. The exposure predictions are coming from those ExpoCast models that we mentioned earlier. We've color coded the exposure predictions for farfield chemicals. These are chemicals that normally don't see residential use versus near-field chemicals in red and those are chemicals that do see residential use. We're indicating that median and the upper 95 percentile so that the dot in the red and in the exposure data, the red and blue plots, the dot is the median and then it goes the bar rising up from it it goes up to the upper 95th percentile of the predicted population variability for exposure. So there's a range of predicted exposures from the median to the upper 95th percentile. And then we're comparing that to the bioactivity data I've mentioned earlier in the talk about the bioactivity data coming from ToxCast from 18 estrogen receptor assays and estrogen receptor model. In this case, we're looking at results that have been converted into something like an AC50. It's called the pseudo ACC here and it's showing the in vitro active concentration at the cut off, the pseudo ACC or pseudo AC50 like value, potency value and showing the median which is the black dot and then the minimum down to 95 percent confidence level. So it's a dot with a line dropping down from it. And what we want to do is compare the exposure predictions to the bioactivity predictions. Remember these have been converted from in vitro to in-vivo. So this is a predicted human bioactivity. And if you go to the right of the diagram you see a lot of separation orders and orders of magnitude separation between estrogen receptor bioactivity and predicted exposure for those chemicals. As we move across, we've arranged the order of chemicals to where you start to see the exposure prediction and the bioactivity prediction becoming closer and closer and then starting to overlap in different ranges, overlap between predicted exposure and predicted bioactivity. Based on the IBER approach, the integrated bioactivity exposure ranking, chemicals less separation between bioactivity and exposure or overlapping bioactive and exposure would be the highest priority chemicals for further screening and testing. This changes the results if we integrate not just bioactivity, but exposure. It changes the ranking or prioritization of chemicals somewhat. It's not just the highest bioactivity chemicals but it's a combination of bioactivity and higher predicted exposure that makes something a greater priority for further screening and testing. So in closing, the future directions for the Endocrine disruptive screening program are to continue to screen these thousands of chemicals into the program using high-throughput and predictive models as an alternative to the low-throughput and animal intensive approaches to develop and validate not just the estrogen, but also the androgen and thyroid high-throughput screening alternatives. To put all of these data and screening and testing high-throughput and more traditional into the context of toxicity pathways and toxicity pathways leading to adverse outcome pathways and as possible to incorporate not just bioactivity data, but exposure data for further chemical prioritization and eventual screening. The combination of bioactivity and exposure has the potential to support high-throughput risk assessments in the future moving beyond just screening and testing. So with that, thank you for your attention. I hope this talk was helpful and I look forward to continuing this partnership with JHU. Thank you.
