Annihilating PFAS with an AI-designed enzyme

1 day ago 2

On a cool Saturday morning in October last year, high school students along with their parents and friends gathered in a brightly lit lecture hall on the California Institute of Technology (Caltech) campus. They chatted with each other as they filed down rows of long tables looking out onto the blue-hued San Gabriel mountains in the distance. A low hum of anticipation filled the hall.

They were at Caltech for an informal gathering of high school International Genetically Engineered Machine (iGEM) teams, a synthetic biology competition for groups of high school or college students. These high school students made up four different iGEM teams and came to Caltech from all over Southern California to present their research projects to each other. By practicing their presentations and fielding questions from their peers, they were preparing for their presentations in front of the judges at the iGEM Grand Jamboree, which would take place in Paris later that month.

From projects using genetically modified bacteria to try to reduce methane production in landfills to finding new ways to break down microplastics, the students undertook research projects with the goal of using synthetic biology to solve real-world problems.

Students wear white lab coats and purple gloves while posing for a photo in Matt Thompson’s laboratory at the California Institute of Technology.

One group — the PFAS Annihilators — took on a very ambitious project. As the team members took turns presenting their slides, murmurs of “Wow, this is amazing for high school students,” trickled through the audience. They explained how over the course of the summer they developed an artificial intelligence (AI) model to design an enzyme that could degrade per- and polyfluoroalkyl substances (PFAS). Not only did they use their model to design multiple PFAS-degrading enzyme candidates, but they also managed to express the proteins — a challenge for any AI-designed protein, but especially for a PFAS-degrading one.

After the team fielded a few questions from their peers, a parent sitting in the back of the room raised his hand and asked, “When did you feel underestimated?”

Visibly relaxing from their formal presentation stances, the team exchanged knowing glances, and one answered, “Gonna be honest, probably they entire time.”

Forever chemicals

From the non-stick coating on pots and pans to the insides of pizza boxes, PFAS are everywhere. With their heat-resistant, waterproof, and oil-resistant properties, PFAS have been used for all kinds of industrial applications since the 1950s (1).

The molecules get these useful qualities from their structure: long chains of extremely strong and stable carbon-fluorine bonds. The problem is that because these bonds are so strong, PFAS molecules don’t break down easily, which gives them their oft-quoted name: “forever chemicals” (2).

Scientists first discovered that PFAS molecules from industrial sites were finding their way into people’s bodies in the 1970s when they noticed the molecules in the bloodstream of the industrial workers who worked with the fluorochemicals (3). But as governments performed more PFAS testing in the early 2000s, they also detected PFAS in the blood of everyday people around the world (4). With more testing came greater regulations for PFAS production. For example, from 2000 to 2002, 3M voluntarily phased out their production of specific kinds of PFAS molecules from their production line, but other kinds of PFAS molecules are still produced today. While people can be exposed to PFAS through a variety of avenues, the most significant way is through drinking water.

“Even World War II military sites, back then they used PFAS, and apparently there's a center in the middle of Iowa where these munitions were left and were not used during the war,” said Alec Lourenco, a graduate student in Matt Thomson’s lab at Caltech as well as the mentor to the PFAS Annihilators iGEM team. “They're actually just sitting there and leaking into the surrounding environment, so being able to bioremediate that would be great.”

Scientists have linked exposure to PFAS with the development of kidney and testicular cancers, metabolic changes like increased cholesterol levels and altered liver enzyme function, and reproductive health complications such as preeclampsia and lower birth weight (5). Researchers have also associated PFAS exposure with a decreased antibody response to vaccines.

“This has been a huge problem in history and just for people in general for decades, so if they're able to identify a PFAS degrader — it would be huge.”
- Alec Lourenco, California Institute of Technology

Normal water treatment facilities can’t remove PFAS, but over the past 20 years or so, there have been an increasing number of industrial efforts to filter PFAS from water supplies (6). While this is a great step, eliminating PFAS altogether would be the ideal. There are many strategies in the works to destroy PFAS — from electrochemical oxidation to incineration — but which method works the best, is the least energy intensive, and the least costly is still not well understood.

“This has been a huge problem in history and just for people in general for decades, so if they're able to identify a PFAS degrader — it would be huge,” said Lourenco.

A PFAS project

The iGEM team that eventually became the PFAS Annihilators started as groups of high school students from all around Los Angeles. Many of them were part of the Solar Energy Activity Lab (SEAL), which is a Caltech outreach program where local high school students work with graduate students and postdoctoral researchers on sustainable energy research projects. Others had been working with Caltech graduate students already, and when they heard about the project, they rushed to join in.

“When you go into this as an undergraduate or high schooler, you kind of expect everyone else to know what's going on, and then you jump into it, and no one really does,” said Troy Reyes, a member of the iGEM team who is now a first year at Pasadena City College. “It's really hard to switch your mindset from kind of a follower to a leader in the field because you need to have confidence in where you're going and especially in breaking new ground.”

At the beginning of 2024, they started their project by diving headfirst into the scientific literature around PFAS and by reaching out to experts in academia and industry to get their expert advice on how best to tackle the PFAS problem.

“We were initially going to create a filter that binds to PFAS, but then, after talking to some stakeholders and also experts like professors … they told us that it would be better to shift our focus towards a PFAS degrader because it would just be more beneficial,” said Willow Nauber, a senior at Westridge School for Girls and a member of the iGEM team.

Ray Chung, a sophomore at EF Academy and a member of the iGEM team added, “We also found that a lot of others already made filtration systems … so we decided that degradation was a lot more progressive.”

One of those experts was Ariel Furst, a chemical engineer at the Massachusetts Institute of Technology. She develops biomaterials to help remediate water from toxic chemicals like pesticides and PFAS. She met with the iGEM team a few times to learn about their project and give them some advice. What struck her most about the students was how engaged they were in understanding the science around PFAS.

“The students are just amazing,” she said. “They always have really insightful and awesome questions.”

The team also spoke with Mamadou Diallo, a chemical engineer with expertise in water treatment at Caltech. He pointed them toward enzymes known as reductive dehalogenases. These enzymes occur naturally in a few specific species of bacteria, and they do one thing really well — break that strong carbon-fluorine bond found in PFAS.

From a large language model to enzymes in the lab

Diallo guided the team to an enzyme from the soil-dwelling microbe, Acidimicrobium sp. strain A6, called A6RdhA (7). A6RdhA naturally degrades PFAS, but the students wanted to see if they could use A6RdhA as a template to create a more efficient PFAS-degrading enzyme using a large language model (LLM).

That was when they hit their first snag — A6RdhA’s protein sequence was incomplete. More than 100 amino acids were missing from its C terminus, which the team worried would make generating new PFAS-degrading enzymes more difficult. To try to fill in the missing amino acids, the students input A6RdhA into Foldseek, which is a resource that searches through protein sequence and structure databases to find proteins with similar features. Doing that in parallel with a literature review, they identified 68 proteins that were structurally similar to A6RdhA.

Two high school students wearing white labcoats and purple gloves press buttons on a PCR machine in a laboratory.

Among those was T7Rdh6, which had the closest structural similarity to A6RdhA. Because of this, the students decided to make a chimera: They filled in A6RdhA’s missing amino acids with those from T7RdhA to create A6T7. From there, they used the 68 proteins they had found as training data for their LLM. Over the course of the next few months, they used their LLM to generate multiple novel AI-designed dehalogenase enzyme sequences.

“The whole fundamental idea of their project is super cool — basically, to use design tools that weren't even available five years ago to design proteins to degrade PFAS,” said Furst.

But the computational work didn’t come without challenges.

“About the AI stuff, a lot of it initially was kind of a time crunch,” said Reyes. “When we were generating the enzymes, it took about, I would say, 30, 20, seconds in order to produce one protein sequence, and we had to do that more than 2,000 times.”

While AI can generate never-before-seen enzymes, there is no guarantee that a cell can produce these new proteins. To conquer this roadblock, the team decided to build an expression classifier: a computational program to predict which AI-created amino acid sequences will express in a cell. They then came up with a mass spectrometry-based readout to test their enzyme classifier.

“I was surprised to find out how hard it was to express AI-generated enzymes,” said Nauber. “It was 42 out of like 8,000.”

Lourenco added, “We're still trying to rule out: Is it operator error, or was it indeed a property of these libraries such that [our PFAS-degrading enzymes are] even less expressible than proteins that people are designing?”

But after months of computational work, the team transitioned from the dry lab to the wet lab to try expressing their new enzymes. Rather than cloning the sequences into Escherichia coli — which is rife with steps where almost anything can and will go wrong — the students used a cell-free expression system to express their proteins.

This, Lourenco explained, is basically “crushed up E. coli cell guts.” Adding in some nucleotides, amino acids, tRNAs, and more to the mix “allows the lysate to do both transcription and then translation,” he added.

The first protein the team set out to express was their chimera: A6T7. But they soon realized that something was missing. Reductive dehalogenases have iron-sulfur clusters as part of their structure. Without these factors in the cell-free expression system, the enzymes couldn’t fold properly. When the team added these clusters into the mix, they finally saw the correctly sized band on the gel that they’d been waiting for — an expressed, folded, and completely novel potential PFAS-degrading enzyme.

“All the proteins they're using are iron-sulfur cluster proteins, which are the most annoying,” said Furst. “Honestly, it's — I don't want to say surprising, because if anybody was going to do it, it was going to be this team — but to get proteins with cofactors actually assembled in them, with these iron-sulfur clusters assembled in them, is like crazy awesome for such a short time.”

The team ended up expressing four additional AI-generated candidate PFAS-degrading enzymes.

Even if they don't reach their overall goal, they will have shown that a bunch of motivated high school students can do real research at a high level and cooperate on it, and so they've expressed these things successfully. And to me, that's a major accomplishment.
– Harry Gray, California Institute of Technology

“The expression work they have done is really first rate,” said Harry Gray, a chemical engineer at Caltech and founder of the SEAL program. “Even if they don't reach their overall goal, they will have shown that a bunch of motivated high school students can do real research at a high level and cooperate on it, and so they've expressed these things successfully. And to me, that's a major accomplishment.”

For the PFAS Annihilators, they were happy to have gotten their enzymes to express. The next step would be to send the enzymes to the Furst lab to test them on actual PFAS, but before that, they needed to prepare for their presentation before the iGEM judges. While the Grand Jamboree competition would be in Paris, the team planned to participate remotely.

“We definitely didn't have enough money to travel,” said Monica Barsever, the team’s Principal Investigator. But then, she got an email from Linda McManus, who together with her husband William Bridges — a Caltech engineering professor emeritus and inventor of the argon ion laser — wanted to donate to help take the team to Paris to compete. “It really fell from the sky at the last minute,” said Barsever. “It was life-changing for us.”

To Paris and beyond

While a couple of the students had been to Paris before, most of them had not, and for some, this trip to Paris would be their first time leaving the country. While they had a chance to do some sightseeing and try their first escargot, the PFAS Annihilators were thrilled to get to be a part of the iGEM competition in person.

“What I liked most was being around everyone, all the iGEMers, because you could feel all of their passion, especially the high schoolers. You could see that everyone was just happy to be there, and they were there for a reason,” said Reyes.

Nauber added, “It was really fun to go up to different people from around the world and talk with them about what they did and ask them how they did their work.”

While they started out feeling a little nervous for the judging portion of the competition, their practice had paid off.

“Once we got into the room with the judges, it just felt like it went really smooth because we knew exactly what we were talking about,” said Nauber.

Their hard work earned them both a Silver Medal and the award for Best Model for their computational work.

“I don't think a lot of us could believe it,” said Reyes. “It was a good surprise.”

Students stand off to the side of a large screen in a lecture hall.

While the competition may have been over, there were still some loose ends to their project that the students wanted to tie up. For one, when the team sent their initial batch of AI-designed enzymes to the Furst lab, they didn’t do so under the right conditions.

“The first batch we got was not sent anaerobically, so I think we ended up losing that cofactor that's important for its activity. But we are working with them to get some fresh cells to do the assays we run to look for activity,” said Furst. “Based on their spectra that they've gotten, it looks like the protein is assembled properly and everything and has this cofactor, so it should be active. Now, we just need to confirm it.”

Furst plans to start by running some simple assays to look at the enzymes’ activities and seeing how well they can break carbon-chlorine bonds, which are easier halogen bonds to break than carbon-fluorine ones. Her team will also perform mass spectrometry and nuclear magnetic resonance (NMR) to see if the enzymes are degrading these compounds.

“It's such a wonderful project, and I think it will stimulate a lot of people out in the drug discovery world to work on things like this. So, my guess is it will be followed up in other labs who will do variations of this,” said Gray. “In that sense, they will have made a huge contribution to sustainable research and a sustainable planet, and for that, I give them enormous credit.”

The team has posted their AI-designed PFAS-degrading enzyme work as a preprint on bioRxiv, and so far, they all seem to be eager to do more research (8).

“Throughout high school, I felt kind of lonely in my endeavors because I couldn't find anyone else who was a science geek, so once I found my people, then it felt really cool. So, for me, I think it still makes me want to do research more than before, and I think I'm always going to want to be doing this kind of stuff,” said Reyes.

Chung added, “I just really like the scientific process in general, being able to collect such large data pools, then process through them,” he said. “I'm definitely going to keep going through science thanks to this.”

Furst is excited to see both where the team’s project goes as well as where the students will go from here.

“Their enthusiasm, it's really infectious,” she said. “When you interact with them, you can see how passionate they are about solving this problem, and that's what we want in future scientists.”

References