Episode 75: Identifying Chemicals in Consumer Products

Lisa Peña (LP): Welcome to a new year of listening and learning. We're starting 2025 with a conversation on chemicals in consumer products. An SwRI led study explored the chemical composition of common household items and the possibility of human exposure. What researchers uncovered and the SwRI tool that illuminated the data, next on this episode of Technology Today. 

[MUSIC PLAYING]
 

We live with technology, science, engineering, and the results of innovative research every day. Now, let's understand it better. You're listening to the Technology Today Podcast presented by Southwest Research Institute. From deep sea to deep space, we develop solutions to benefit humankind. Transcript and photos for this episode and all episodes are available at podcast.swri.org. Share the podcast and hit that subscribe button on your favorite podcast platform.

Hello and welcome to Technology Today. I'm Lisa Pena. SwRI collaborated with the Environmental Protection Agency to study household products, examining their chemical makeup and evaluating the potential for human exposure. They looked at clothing, upholstery, and products made of rubber and plastics to determine if chemicals could be extracted or emitted by normal use. The study was published in the Environmental Science and Technology Journal.

Our guests today are the SwRI researchers who led the study. Analytical chemist Dr. Kristin Favela and research engineer William Watson. They're here to tell us about their findings and an innovative SwRI tool used to analyze the chemicals. Thank you for being here, Kristin and William, and for being our first guests of 2025.

Image

This image shows chemical features captured with SwRI-developed machine-learning tool Highlight™. SwRI collaborated with the Environmental Protection Agency to evaluate the chemical makeup of common household items. The study determined whether the chemicals could be emitted or extracted with normal use and the risk of human exposure.

Kristin Favela (KF): Thank you for having us. 

William Watson (WW):Yeah, thanks for having us. 

LP: All right, so let's start with an overview of this important study. What was the purpose of this research? 

KF: This research is part of an ongoing study or ongoing body of work, I should say, to try to characterize household products and consumer care products. So in this particular study, we had looked at these products in terms of the chemical makeup by extracting them, which means we used a liquid to pull out the chemicals in the products. So kind of like making tea. 

However, there were questions about is that really relevant? Because in normal usage, we aren't pulling the chemicals out of the products with different solvents. So we took a look at those samples, again, using a different design that was intended to collect the vapor signatures. So looking at what is actually emitted from the product. 

So this study was really about what's there and how do those chemicals come out of those products. And it turns out that, of course, vapor signatures may be more relevant to exposure for the reasons I just enumerated, that when we're using these products, we're not going to be pulling chemicals out with a liquid. 

LP: Yeah, and let's put that into perspective. So we're sitting on a couch. We're not using, as you said, a solvent to extract chemicals from a couch. So you're looking at just what is emitted through normal use, let's say sitting. 

KF: That's right. And that's an important point, because analytically, it's easier to do the former, to characterize the product by extracting it with a liquid. But then questions arise. Is that really fair? So in this study, we did both and compared them. 

WW: I can also add that another unique challenge of this research was that the data was collected over the course of several years, which posed a unique data processing challenge, which is where we could use some of our innovative machine learning tools. 

LP: So we're going to get into those tools in a bit. We're excited to learn about SwRI tools to find these chemicals. But let's understand how can exposure to chemicals in consumer products impact human health? It sounds like something you want to avoid, chemicals in my household products. But you know what are possible effects? 

KF: Well, taking a step back, it turns out that most consumer products are not even characterized. So that makes the problem more difficult. How can you even begin to predict human health impacts if you don't know what is present in the products? And our first manuscript in this arena we published with the EPA in 2018 demonstrated that roughly 80% of the chemicals we found in the household products and consumer care products that we examined were not previously known to be in consumer products. 

So this body of work and this study is really about trying to solve the first piece of the puzzle, what is there. And then EPA takes our data and they're attempting to build models to answer the second question about human health impacts using their ToxCast program. 

LP: So two questions there. What is there? What is in the product? And then what is the impact to human health? SwRI's role, from what I'm understanding, is to answer that first question. What is there? And then pass that information onto the EPA so then they can determine and what is the impact to human health? 

KF: Yes, that's right. And it's important to understand that the techniques that we're using are untargeted. And there are, in some cases, hundreds of chemicals in each product. So that makes this work extraordinarily complex and couldn't be done without the modern machine learning tools that we have. 

LP: And when you say untargeted, could you define that? 

KF: Yes. Untargeted means we're trying to characterize everything that is detected by the instrument, as opposed to looking for only specific things, which is largely what was done in the past. 

LP: So what type of items did you investigate? I know we talked about rubber, plastic materials, and we said clothing, upholstery. But can you name specific items? Did you look at jeans or water bottles? What type of household or personal items are we talking about? 
 

Image

A single sample can yield 100s to 1000s of chemical signals. You can read more about the study, “Discerning Emittable from Extractable Chemicals Identified in Consumer Products by Non-Targeted GCxGC-TOFMS,” in the Environmental Science & Technology journal.

KF: So there were two main categories. The first was clothing, fabric, and upholstery. So that includes things like baby socks and-- 

WW: Sleepwear. 

KF: Sleepwear, t-shirts. Also upholstery that you may use to outfit a couch. And then on the rubber and plastic, these were things like rubber play mats that children may play on. Things of that nature. 

LP: Like the gym flooring. 

KF: Yes. 

LP: Okay. So you're looking at these items and I think all of our antenna go up. Like, OK, I wear these things. I have all these things. And so I'd like to say up front, so you looked at these items, and we don't necessarily have a determination on the possibility of human exposure to chemicals from these items. But we're going to talk about what you found. 

The study looked at chemicals present in household items. That was one piece of the puzzle. The next part, as we discussed, was whether the chemicals could be emitted or extracted, again, with normal use. Now, if you could explain to us a little bit further why is this part key? Emitted or extracted with normal use.

KF: Yeah, so vapor signatures may be more relevant to exposure because they come out of the product without being forced out. And as we discussed, forcing them out with a solvent is analytically easier but may not imitate real life. So we looked at two different temperatures with the emission work. One was approximately human body temperature and the other was approximately the temperature of a hot car. 

LP: And so when you say these vapor signatures, does that mean the-- well, can you explain more what that means? Does that mean that it would be breathed in? We would breathe that in potentially if it's emitted? 

KF: Potentially. So the way the experiment was performed was that a small portion of the product was put into a jar with a media that can collect the chemicals that are emitted into the air. And then that media is tested with the instrument. So these are chemicals that were present in that headspace. 

LP: Okay, let's dig a little deeper into the tools and methods used to conduct the study. How did you determine what chemicals were present in these items? You've gone over it a little bit. But if we could get a little bit more in depth, again, whether they could be emitted or extracted, and human exposure. What type of methods and tools are we talking about? 

KF: Well, first of all, a shout out to our extraction lab, which is a huge part of this study of probably, I would argue, the most important part, because this is where the samples are prepared. And our dedicated staff in the extraction lab, Christina Menn and the now retired Hamid Idrissi, have been doing this kind of work for years and have developed some really wonderful techniques to perform these sample preparation techniques, which is a big challenge, because we are doing trace analysis. So they have to be very careful, for example, to not contaminate the samples and things of that nature. 

So after they prepared the samples, then we analyzed them using two dimensional gas chromatography time of flight mass spectrometry. And we, as William alluded to, ended up with about four years of data to process and well over 300 files, each file containing potentially certainly hundreds but potentially thousands of signatures. So this was a major data processing problem. So how do you process all that data? And this is where computer science plays a major role. 

LP: Okay. So I want to review this extraction process again. So it starts with a product, or let's say, as you mentioned, the baby socks. So they prepare-- can you walk us through again what you mean by they prepare the sample? 

KF: Sure. So again, we're doing extraction and emission. So for extraction, what they would do would be to take a small portion of the sample, maybe cut it into even smaller pieces, and then add that product to a solvent. And then the solvent would be taken out and analyzed on the instrument. So it's kind of making tea. 

On the emission side, what they would do would be, again, to take the same amount of the same product, put it in a jar, a closed jar, with a sorbent that's capable of absorbing chemicals from the air. And they would heat it to the correct temperature for the designated amount of time. And then they would take that sorbent out that has all of the absorbed chemicals from the headspace, and then they would extract that sorbent with a solvent. And then that's analyzed on the instrument. 

So at the end of the day, we're analyzing a liquid. But did that liquid have direct contact with the sample or are we extracting what was pulled out of the air? So that's the difference between a liquid extraction and the emission experiment. 

LP: Okay. And you mentioned four years of data, 300 files, and so many, you mentioned the term signatures. Can you explain what you mean by signatures? 
 

Image

Dr. Kristin Favela, SwRI analytical chemist and study co-author, analyzed how samples of rubber, plastic, clothing, upholstery and fabric responded to environmental factors, such as heat, and solvents, such as water. The collaborative study is published in the Environmental Science & Technology Journal.

KF: So we tend to use the term signatures and features, compounds, chemicals, chemical groups interchangeably. But basically, they're all essentially referring to a chemical in a sample. 

LP: Kristin, this is not your first time on the podcast. We learned about Floodlight and Searchlight in March 2021 on episode 29. So now you are working with a new tool, Highlight. So how does highlight compare to Floodlight and Searchlight? 

WW: Yeah, so Floodlight and Searchlight were tools developed using internal research R&D funds back in 2020. And they won an R&D 100 award in 2021. So Highlight is the next generation of those tools. So Floodlight and Searchlight were built using a variety of different supervised machine learning techniques, trained on around 70,000 manually labeled examples and incorporating other techniques to perform sample alignment or aggregation of different sample files into group compounds across samples. But they were focused more on low resolution data. 

So Highlight, the next generation, focuses on what we call high resolution data, which is a more complex problem. So to build this tool, we used transfer learning, which is a technique where we can take the models that were trained for low resolution data and then essentially just fine tune them, retrain them using a smaller set of data. In this case, around 13,000 new manually labeled examples. 

And then we also incorporated some new iterative processing techniques to allow us to pull out lower level signatures and a new machine learning model as well to predict the identification confidence. So all of these chemical signatures are assigned an ID about what they might be. And that ID could be totally wrong. It could be very confident. So we developed a model trained on around 200,000 examples to predict that identification confidence. And then based on the settings used in the program, we estimate that you can get anywhere between 10 times to 100 times speedup in the processing workflow. 

LP: Can you connect the dots for us from extraction lab to Highlight? What happens with that sample in between those two, or how does it go from one to the other? 

WW: Okay, so once the sample is prepared in the extraction lab, it is then analyzed on the instrument, which Kristin typically runs. And then the data from there goes through a little bit of manual processing first before it is put through our machine learning tools. And in those machine learning tools, Kristin can also go through and review the processing that we've performed. And then at that point, the data is in its final state and ready for being passed off to the EPA. 

KF: Yeah, and that's really important, because before we had this tool, we would give the EPA essentially a laundry list of what was found in every sample as an Excel spreadsheet. And it was really difficult to make connections between samples. So without this tool, considering that this is four years worth of data, this work absolutely would not be possible. 

And that's because every time you have a session at the instrument, things are slightly different. What we call the retention time, which is how long a given chemical's retained on the instrument, that retention time drifts slightly over time. So without these kinds of tools, it would have been completely, I would say, impossible, or at least the amount of time it would take to make all these connections between all these samples would have been really, really difficult. 

LP: I remember in our conversation on Floodlight and Searchlight the discussion on freeing up man hours, freeing up manpower, and these tools taking over some of that work so our analytical chemists and our engineers and our computer scientists can focus on the work at hand instead of sifting through all the data. So essentially, Highlight is doing this even faster now. 

KF: Yeah, absolutely, absolutely. And that's correct. So Floodlight and Searchlight were designed to reduce that burden of all the individual manual signal review. Highlight retains all of that functionality. But it was designed for high resolution mass spec data. So as William was talking about, it adds this iterative step that really leverages the high resolution nature of the data collection. 

LP: And William, maybe you can add on why is high resolution so important in this work? 
 

Image

William Watson, research engineer in SwRI’s Intelligent Systems Division, pictured left at the board, was the lead author of the study. Watson explains, once product samples were processed through the Institute’s extraction lab, chemicals present in the items were analyzed with SwRI’s Highlight™ software. Highlight leverages machine learning algorithms for rapid pattern matching, accelerating workflow.

WW: So the difference between high and low resolution is that so low resolution data, the information that's coming off of the instrument, the mass spectrometric data, is all integer mass. So it's telling you just what the mass is with no decimal places, which does not provide you with as much information as if you had those places after the decimal, which is what high resolution data provides, which allows you to be much more specific. So when we go through the iterative process where we're looking for further data, which we call the fishing step, we can use that very specific mass as kind of a fingerprint to allow us to identify with great precision what is present in just the raw soup of data. 

LP: Okay, so we've heard about the study. We've heard about your tools. But let's get into your findings I think we're all curious about. Did you find chemicals being extracted or emitted from household products? Were they being emitted, extracted at dangerous levels? And do we know the actual impact to human health at this point? 

KF: Well, again, our main goal with this study is cataloging, not predicting impact. But we did find that temperature has a huge impact on the amount of material emitted, and that extracting the samples with the liquid or the solvent pulls out even more than emissions. So that's really good news for health. So it's pointing to the fact that when we do our liquid extraction, that's worst case scenario. So all those chemicals are present, but we're forcing them out. But much less of that mass actually gets forced out when the sample is just sitting there emitting chemicals. And of course, the higher temperatures force more chemicals out than the lower temperatures. 

LP: So were you able to determine which items were most risky? 

KF: Well, we did find some interesting chemicals. Antioxidants such as six PPD and BKF. We also found an antifungal agent, dichlorine. We found phenol, BHT, which is butylated hydroxytoluene, phthalates, et cetera. All of these chemicals, they may be biologically active, but their effects need to be investigated further. 

LP: You just listed you had a long list of chemicals that you found. But what does that mean in terms of products or types of products? Did you find a type of product or clothing or upholstery that you think was most risky? 

WW: In general, the rubber products were riskier than the fabrics, clothing, upholstery. So things like rubber playmats or tires, those kind of things tend to have more possibly unpleasant chemicals in them. 

KF: Yeah, that's correct. So some of the products were recycled tire products that get used on playgrounds and things like that. And those did have a lot more chemicals in them. And those materials are also known to be sort of absorbent. So they pull things out of the environment and then may re-emit them. 

LP: Oh, okay. Yeah, that's interesting. But again, this was finding those chemicals and figuring out whether or not they could be emitted or extracted, but not necessarily what their impact is on human health yet. 

KF: Correct. So all of this data is going into EPA's models to answer that really important question about human health. Again, if you don't know it's there-- no one has unlimited resources. So there needs to be a sort of triage process. So what chemicals should we be most concerned about? What chemicals have the most potential for being toxic and/or are present at highest levels? 

LP: So did you have a most surprising finding or outcome from the data that you extracted? 

WW: I would say that most of the stuff that we or the findings that we had were kind of in line with what we would have assumed going in. So we found that liquid extraction pulls out more than high temperature emission, which pulls out more than low temperature emission, which kind of makes sense, just on the face of it, that if you're forcing something out with liquid extraction, they'll get the most. At low temperature, you're not going to have many things emitting. But no one had actually done this study before, so everything had just been based kind of on that assumption. So now we've kind of provided a foundation for further modeling. 

LP: You found that in normal wear and tear, a lot of these chemicals are staying put. Is that accurate? 

KF: That's correct. Yeah. 

LP: So ultimately, your answer was not really any surprises. 

WW: No, not really. It was in line with what we had expected. But again, we're actually providing evidence for future work. So it's not just based on conjecture. 

LP: Is there a specific item or a type of material or fabric that you no longer use in your home due to your research? 

KF: Well, since the health effects of most of these compounds are still being investigated, there's not yet a clear linkage between the use of any specific products in health. So to be honest, I try not to think about it too much when I'm at the grocery store buying products. 

But I will say we have a previous study published with the EPA looking at chemicals of different cohorts in blood samples. And one of the takeaways of that study was that women tend to have more chemicals from household products in their system than men. So I would advise all the women out there to know that you're beautiful just the way you are, and maybe reduce some of the consumer care products that are not absolutely necessary. 

LP: Right. Because we are doing the creams and the makeup and all the body washes and lotions and potions. 

KF: Correct. 

LP:That's interesting. 

KF:Yeah. 

LP: Maybe dial that back a bit. All right. 

WW: I would agree with Kristin. I really try not to think about it too much, because there's only so much you can do. You can't avoid exposure to fabric or something. 

LP: When you're talking about as much data as you collect, I can see where that information would just be overwhelming if you don't really know what the final results are, and you try to work that into your daily life. 

KF: And not all the chemicals are bad. So we found, for example, in this study, I believe we found oleic acid, which is known to be positive for human health. So just because it's present in a product doesn't mean it's necessarily bad. 

LP: And the word chemical doesn't necessarily mean bad or harmful. 

KF: That's right. 

LP: OK. So let's discuss the field of exposomics. Did I say that correctly? 

KF: Yes. 

LP: The field of exposomics. Advancing this field is also part of the study. So what is exposomics and how is this work contributing to the field? 

WW: So exposomics is the study of the exposome. So what is the exposome? The exposome is a person's total environmental exposure and the related health effects throughout their lifetime. So from your birth to your death, it includes all kinds of environmental exposures, be it from the air, from consumer products, from anything. And then how those exposures interact with your health, your biological systems. So this study is focusing more on the first part of that, informing the actual environmental exposure. So what are you actually exposed to in your day to day life? 

LP: Would you call it a new field? 

KF: I don't think it's a new field, because I think that targeted analysis has been performed for quite some time. So you could think of measuring lead as a targeted analysis. So we lead is bad. So let's look for lead. That kind of work's been ongoing for a really long time. 

I think what is really supercharging this field right now is the untargeted analysis. So instead of just looking for specific things, let's try to do a full characterization of what's present. That's really what's new. And that's been enabled by recent advances in instrumentation and especially computer science.

LP: So going back to your study again, published in the Environmental Science and Technology Journal, we'll have a link on this episode page. Is this study ongoing and what do you hope to achieve with your research in 2025? 

KF: So the next step is to use this study to build a model for interpreting extracted data, which is easier to collect. So one of the goals of this study was to, again, to compare the extracted chemicals to the emitted chemicals. So now that we have that data and we were able to make those comparisons, we want to use that as a model so we can do the easier experiment, which is extracting with a solvent, and then apply our model to know what is most likely to be admitted. 

We also want to leverage signatures that may be found in hair to predict health outcomes and inform exposure risks. And that's because many external exposures actually seep into your hair from the outside and become trapped. So we think that that could be a good reservoir for predicting human health outcomes based on what you've actually been exposed to. 

LP: And so part of that would be understanding first the chemical, the impact the chemicals have on human health. And then once finding or identifying those harmful chemicals in a sample of hair, then understanding that the person with those chemicals in their hair sample could experience those effects potentially. 

KF: Correct. So that's an avenue of research that we're going down that would be. That's independent of our work with the EPA. 

WW: And I think also that in the future, so 2025 and beyond, tools like Highlight or other software tools that are capable of dealing with large amounts of data are going to be increasingly used to look retrospectively at data that's been collected in the past. For example, with the study over four years of data and hundreds of files. 

But other studies like that that have been conducted in the past but have not been fully examined simply due to the fact that it would have taken too long. There's a labor limitation in the past. And now that we have these new tools, we can go back and re-examine and perhaps draw better conclusions, get more out of that data that's already been collected. 

LP: It's like DNA technology nowadays, taking those old cold cases and running them through current DNA technology and finding new links. So kind of that same thought process. 

WW: Yeah, it's a good analogy. 

LP: Do you think the information that you've found through your work, do you think that it could eventually change the way products are made? 

KF: Yes, I think potentially, and I think to be fair to the manufacturers, I don't think most manufacturers know that all these chemicals are present in their products, because it does take right now a certain amount of expertise to look at that. And they would have to be examining all of their intermediates. And they're dependent on whatever their suppliers give them. 

And especially during the COVID pandemic, there was a lot of disruptions in the supply chain. And we did see some unusual things during that time period. Product compositions changing. And so I don't think that the manufacturers are doing this on purpose. I just I don't think they're aware of everything that's in their products. So if we can give them-- if we can provide a tool that allows maybe manufacturers to screen their intermediates and their supply chain a little bit more easily and inexpensively, then that could have a positive impact on reducing certain chemicals in the end product. 

LP: So again, going back to that baby sock example. They're sourcing this material or the cotton from a supplier, and maybe not knowing what's happening along the chain. And as you said, there may be chemicals in there that they have no clue about. 

KF: Correct. 

LP: Just depending on where the cotton was grown and-- 

KF: Yeah, absolutely. 

LP: Tons of other factors. OK. All right. So I always like to ask, how does your research and this newest tool, Highlight, fulfill SwRI's mission to engage in work that benefits humankind? 

WW: So Highlight, which we developed internally through our R&D program, provides researchers with the tools they need to more fully interrogate our environment, understand what chemical exposures there are, and then better understand the consequences of those chemical exposures. 

KF: I just want to give a shout out to Michael Hartnett and Jake Janssen who were very instrumental in helping us build the Floodlight, Searchlight, and Highlight products, and that's really enabling-- those tools are really what's enabling the future. 

LP: All right, I love that. Enabling the future. Again, the study was published in the Environmental Science and Technology Journal, a collaboration between Southwest Research Institute and the EPA. Kristin and William, thank you for being here. And again, for helping us understand this important study, your research is certainly contributing to a safer world, and I wish you all the best as you move forward with your work in 2025 and beyond. 

KF: Thank you. 

WW: Yes, thank you so much for having us. 

 

And thank you to our listeners for learning along with us today. You can hear all of our Technology Today episodes, and see photos, and complete transcripts at podcast.swri.org. Remember to share our podcast and subscribe on your favorite podcast platform.

Want to see what else we're up to? Connect with Southwest Research Institute on Facebook, Instagram, X, LinkedIn, and YouTube. Check out the Technology Today Magazine at technologytoday.swri.org. And now is a great time to become an SwRI problem solver. Visit our career page at SwRI.jobs.

Ian McKinney and Bryan Ortiz are the podcast audio engineers and editors. I am producer and host, Lisa Peña.

Thanks for listening.

[MUSIC PLAYING]