Jonathan Choi is a chemicals policy fellow.
EDF Senior Scientist Dr. Jennifer McPartland contributed to this post.
The beginning of this century will no doubt be known for a lot of things. In the biological sciences, I predict it’ll be known for big data. It’s hard to wrap your head around just how far we’ve come already. For example, the data chips that sing “happy birthday” to your loved ones in those horrendously overpriced cards have more computing power than the Allies did in 1945. When I first started using computers, the 5.4” floppy disk was being replaced by the new 256Kb 3.5” disk. Now in Korea, you can get 1 GB per second internet speeds for $20 a month. That’s around 4000 floppy disks of data per second for about as much as I spend every week at the burrito place down the street.
In the biological sciences, we’ve seen an explosion of new ways to generate, collect, analyze, and store data. We’re photographing the world’s biodiversity and sharing it with crowdsourced taxonomists. We’re creating a database of the genomes of the world’s organisms. We’re mapping chemical exposures (our exposome), inventorying the microbes that live in our guts (our microbiome), ripping apart cells and sequencing every bit of messenger RNA that floats around inside (our transcriptome), and much more.
So, it’s not too surprising that regulatory agencies like EPA are pushing their own efforts to amass large quantities of data to help meet their missions. EPA has the unenviable task of reviewing tens of thousands of chemicals currently on the market with little health and safety data, on top of hundreds of new chemicals banging at its door each year. As we have written on numerous occasions, the agency clearly needs a better law that gives it greater authority to get the data it needs to effectively evaluate and manage chemical risks. But, given the information abyss in which we operate, we could definitely stand to adopt new testing approaches that generate at least screening level data on chemicals faster.
Toward this end, EPA has developed a program called ToxCast (short for Toxicity foreCaster) which we’ve discussed before. In short, ToxCast is a high-throughput screening program that couples automated testing of 1000s of chemicals across 100s of tests (or assays) with powerful data processing and management capabilities. It presents a remarkable opportunity to generate information on many chemicals in a fraction of the time typical of traditional toxicity testing and in a manner that has the potential to provide deeper biological insight. The aim, at least initially, is to predict whether a chemical may lead or contribute to a negative health effect and therefore help prioritize where the agency should focus its attention. A good thing!
While this is all very exciting, it’s vitally important to continuously examine ToxCast to identify its strengths as well as weaknesses so that 1) we understand what types of decisions can and can’t be made using the data and 2) we can identify opportunities to make the program stronger.
That’s why a recent article in Environmental Health Perspectives caught our eye. A team of researchers led by investigators at the University of California-Irvine examined 21 chemicals that ToxCast indicated interact with a protein receptor called PPARγ, which, among other metabolic processes, is involved in adipogenesis–the development of fat cells from stem cells. More specifically, the researchers evaluated these 21 chemicals using their own set of tests to identify chemical interactions with the PPARγ receptor. In their own tests, 5 of the 21 chemicals stimulated PPARγ activity, 3 blocked the activity of PPARγ, and the rest were inactive.
In a separate evaluation, the researchers sought to compare the output of a ToxCast-based prediction model (ToxPi, short for Toxicology Priority Index) for adipogenesis that the investigators developed. Only 7 of the 17 chemicals that ToxPi predicted to be adipogenic were positive in the researchers’ own assays. At the same time, 2 of the 7 ToxPi compounds predicted to be negative for adipogenesis were positive in their assays. When the investigators re-ran this evaluation using a refinement of the data designed to weed out false positive chemicals all of the false positives fell off. But, so did all of the true positives.
There were other curious observations. For example, the researchers found a lack of agreement across different ToxCast assays that tested for the same endpoint (e.g., PPARγ activation), with some yielding positive and others yielding negative results. When they used ToxCast assays to look at RXR, a different receptor involved in adipogenesis, they found chemicals that activated one variant (RXRβ), but not another (RXRα). The authors suggested that this was biologically implausible, because there are no compounds known to bind to one but not the other receptor variant.
Obviously, the findings raise some questions, and the study authors provide a couple possible reasons for the disparate correlations they observed. They suggest that the ToxCast assays currently available for examining adipogenesis may not be the best for the job. They also suggest a need to identify and cull out poorly performing assays and identify and add higher-performing screens to the testing battery.
Is the whole ToxCast effort a waste of time, then? No. The authors note that ToxCast shows more predictive ability in other areas of biology. But the authors’ perspective is that it appears to be seriously underperforming in identifying compounds that could interfere with normal adipogenesis.
For its part, the agency seems open to improving its testing methodologies. For example, EPA just launched a new challenge to better reflect human metabolism in its testing program. We believe that this initiative is important, given that, for a number of compounds, it’s what the chemical breaks down into and not the chemical itself that drives toxicity.
At the bigger picture level, it’s fantastic to see a group of university researchers exploring the ToxCast data. As EDF has advocated for years, it is essential that this type of engagement and collaboration with the broader academic research community occur. Such engagement represents one of the most effective ways to advance predictive toxicology and simultaneously build confidence in the agency’s efforts. Here at EDF we look forward to a future of new chemical testing approaches that ultimately work to protect human health.
(Oh one more note: A former EDFer thought to write a blog of this study as well. You can check it out here: http://bit.ly/1PZAmIl).